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_ElecEwSw_VdwNone_GeomW3W3_VF_sse2_single
38 * Electrostatics interaction: Ewald
39 * VdW interaction: None
40 * Geometry: Water3-Water3
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecEwSw_VdwNone_GeomW3W3_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 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
75 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
76 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
77 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
78 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
79 __m128 dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
80 __m128 dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
81 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
82 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
83 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
84 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
85 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
86 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
87 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
90 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
92 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
93 real rswitch_scalar,d_scalar;
94 __m128 dummy_mask,cutoff_mask;
95 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
96 __m128 one = _mm_set1_ps(1.0);
97 __m128 two = _mm_set1_ps(2.0);
103 jindex = nlist->jindex;
105 shiftidx = nlist->shift;
107 shiftvec = fr->shift_vec[0];
108 fshift = fr->fshift[0];
109 facel = _mm_set1_ps(fr->epsfac);
110 charge = mdatoms->chargeA;
112 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
113 ewtab = fr->ic->tabq_coul_FDV0;
114 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
115 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
117 /* Setup water-specific parameters */
118 inr = nlist->iinr[0];
119 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
120 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
121 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
123 jq0 = _mm_set1_ps(charge[inr+0]);
124 jq1 = _mm_set1_ps(charge[inr+1]);
125 jq2 = _mm_set1_ps(charge[inr+2]);
126 qq00 = _mm_mul_ps(iq0,jq0);
127 qq01 = _mm_mul_ps(iq0,jq1);
128 qq02 = _mm_mul_ps(iq0,jq2);
129 qq10 = _mm_mul_ps(iq1,jq0);
130 qq11 = _mm_mul_ps(iq1,jq1);
131 qq12 = _mm_mul_ps(iq1,jq2);
132 qq20 = _mm_mul_ps(iq2,jq0);
133 qq21 = _mm_mul_ps(iq2,jq1);
134 qq22 = _mm_mul_ps(iq2,jq2);
136 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
137 rcutoff_scalar = fr->rcoulomb;
138 rcutoff = _mm_set1_ps(rcutoff_scalar);
139 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
141 rswitch_scalar = fr->rcoulomb_switch;
142 rswitch = _mm_set1_ps(rswitch_scalar);
143 /* Setup switch parameters */
144 d_scalar = rcutoff_scalar-rswitch_scalar;
145 d = _mm_set1_ps(d_scalar);
146 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
147 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
148 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
149 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
150 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
151 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
153 /* Avoid stupid compiler warnings */
154 jnrA = jnrB = jnrC = jnrD = 0;
163 /* Start outer loop over neighborlists */
164 for(iidx=0; iidx<nri; iidx++)
166 /* Load shift vector for this list */
167 i_shift_offset = DIM*shiftidx[iidx];
168 shX = shiftvec[i_shift_offset+XX];
169 shY = shiftvec[i_shift_offset+YY];
170 shZ = shiftvec[i_shift_offset+ZZ];
172 /* Load limits for loop over neighbors */
173 j_index_start = jindex[iidx];
174 j_index_end = jindex[iidx+1];
176 /* Get outer coordinate index */
178 i_coord_offset = DIM*inr;
180 /* Load i particle coords and add shift vector */
181 ix0 = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
182 iy0 = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
183 iz0 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
184 ix1 = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
185 iy1 = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
186 iz1 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
187 ix2 = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
188 iy2 = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
189 iz2 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
191 fix0 = _mm_setzero_ps();
192 fiy0 = _mm_setzero_ps();
193 fiz0 = _mm_setzero_ps();
194 fix1 = _mm_setzero_ps();
195 fiy1 = _mm_setzero_ps();
196 fiz1 = _mm_setzero_ps();
197 fix2 = _mm_setzero_ps();
198 fiy2 = _mm_setzero_ps();
199 fiz2 = _mm_setzero_ps();
201 /* Reset potential sums */
202 velecsum = _mm_setzero_ps();
204 /* Start inner kernel loop */
205 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
208 /* Get j neighbor index, and coordinate index */
214 j_coord_offsetA = DIM*jnrA;
215 j_coord_offsetB = DIM*jnrB;
216 j_coord_offsetC = DIM*jnrC;
217 j_coord_offsetD = DIM*jnrD;
219 /* load j atom coordinates */
220 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
221 x+j_coord_offsetC,x+j_coord_offsetD,
222 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
224 /* Calculate displacement vector */
225 dx00 = _mm_sub_ps(ix0,jx0);
226 dy00 = _mm_sub_ps(iy0,jy0);
227 dz00 = _mm_sub_ps(iz0,jz0);
228 dx01 = _mm_sub_ps(ix0,jx1);
229 dy01 = _mm_sub_ps(iy0,jy1);
230 dz01 = _mm_sub_ps(iz0,jz1);
231 dx02 = _mm_sub_ps(ix0,jx2);
232 dy02 = _mm_sub_ps(iy0,jy2);
233 dz02 = _mm_sub_ps(iz0,jz2);
234 dx10 = _mm_sub_ps(ix1,jx0);
235 dy10 = _mm_sub_ps(iy1,jy0);
236 dz10 = _mm_sub_ps(iz1,jz0);
237 dx11 = _mm_sub_ps(ix1,jx1);
238 dy11 = _mm_sub_ps(iy1,jy1);
239 dz11 = _mm_sub_ps(iz1,jz1);
240 dx12 = _mm_sub_ps(ix1,jx2);
241 dy12 = _mm_sub_ps(iy1,jy2);
242 dz12 = _mm_sub_ps(iz1,jz2);
243 dx20 = _mm_sub_ps(ix2,jx0);
244 dy20 = _mm_sub_ps(iy2,jy0);
245 dz20 = _mm_sub_ps(iz2,jz0);
246 dx21 = _mm_sub_ps(ix2,jx1);
247 dy21 = _mm_sub_ps(iy2,jy1);
248 dz21 = _mm_sub_ps(iz2,jz1);
249 dx22 = _mm_sub_ps(ix2,jx2);
250 dy22 = _mm_sub_ps(iy2,jy2);
251 dz22 = _mm_sub_ps(iz2,jz2);
253 /* Calculate squared distance and things based on it */
254 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
255 rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
256 rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
257 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
258 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
259 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
260 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
261 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
262 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
264 rinv00 = gmx_mm_invsqrt_ps(rsq00);
265 rinv01 = gmx_mm_invsqrt_ps(rsq01);
266 rinv02 = gmx_mm_invsqrt_ps(rsq02);
267 rinv10 = gmx_mm_invsqrt_ps(rsq10);
268 rinv11 = gmx_mm_invsqrt_ps(rsq11);
269 rinv12 = gmx_mm_invsqrt_ps(rsq12);
270 rinv20 = gmx_mm_invsqrt_ps(rsq20);
271 rinv21 = gmx_mm_invsqrt_ps(rsq21);
272 rinv22 = gmx_mm_invsqrt_ps(rsq22);
274 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
275 rinvsq01 = _mm_mul_ps(rinv01,rinv01);
276 rinvsq02 = _mm_mul_ps(rinv02,rinv02);
277 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
278 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
279 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
280 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
281 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
282 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
284 fjx0 = _mm_setzero_ps();
285 fjy0 = _mm_setzero_ps();
286 fjz0 = _mm_setzero_ps();
287 fjx1 = _mm_setzero_ps();
288 fjy1 = _mm_setzero_ps();
289 fjz1 = _mm_setzero_ps();
290 fjx2 = _mm_setzero_ps();
291 fjy2 = _mm_setzero_ps();
292 fjz2 = _mm_setzero_ps();
294 /**************************
295 * CALCULATE INTERACTIONS *
296 **************************/
298 if (gmx_mm_any_lt(rsq00,rcutoff2))
301 r00 = _mm_mul_ps(rsq00,rinv00);
303 /* EWALD ELECTROSTATICS */
305 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
306 ewrt = _mm_mul_ps(r00,ewtabscale);
307 ewitab = _mm_cvttps_epi32(ewrt);
308 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
309 ewitab = _mm_slli_epi32(ewitab,2);
310 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
311 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
312 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
313 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
314 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
315 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
316 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
317 velec = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
318 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
320 d = _mm_sub_ps(r00,rswitch);
321 d = _mm_max_ps(d,_mm_setzero_ps());
322 d2 = _mm_mul_ps(d,d);
323 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
325 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
327 /* Evaluate switch function */
328 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
329 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
330 velec = _mm_mul_ps(velec,sw);
331 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
333 /* Update potential sum for this i atom from the interaction with this j atom. */
334 velec = _mm_and_ps(velec,cutoff_mask);
335 velecsum = _mm_add_ps(velecsum,velec);
339 fscal = _mm_and_ps(fscal,cutoff_mask);
341 /* Calculate temporary vectorial force */
342 tx = _mm_mul_ps(fscal,dx00);
343 ty = _mm_mul_ps(fscal,dy00);
344 tz = _mm_mul_ps(fscal,dz00);
346 /* Update vectorial force */
347 fix0 = _mm_add_ps(fix0,tx);
348 fiy0 = _mm_add_ps(fiy0,ty);
349 fiz0 = _mm_add_ps(fiz0,tz);
351 fjx0 = _mm_add_ps(fjx0,tx);
352 fjy0 = _mm_add_ps(fjy0,ty);
353 fjz0 = _mm_add_ps(fjz0,tz);
357 /**************************
358 * CALCULATE INTERACTIONS *
359 **************************/
361 if (gmx_mm_any_lt(rsq01,rcutoff2))
364 r01 = _mm_mul_ps(rsq01,rinv01);
366 /* EWALD ELECTROSTATICS */
368 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
369 ewrt = _mm_mul_ps(r01,ewtabscale);
370 ewitab = _mm_cvttps_epi32(ewrt);
371 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
372 ewitab = _mm_slli_epi32(ewitab,2);
373 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
374 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
375 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
376 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
377 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
378 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
379 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
380 velec = _mm_mul_ps(qq01,_mm_sub_ps(rinv01,velec));
381 felec = _mm_mul_ps(_mm_mul_ps(qq01,rinv01),_mm_sub_ps(rinvsq01,felec));
383 d = _mm_sub_ps(r01,rswitch);
384 d = _mm_max_ps(d,_mm_setzero_ps());
385 d2 = _mm_mul_ps(d,d);
386 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
388 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
390 /* Evaluate switch function */
391 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
392 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv01,_mm_mul_ps(velec,dsw)) );
393 velec = _mm_mul_ps(velec,sw);
394 cutoff_mask = _mm_cmplt_ps(rsq01,rcutoff2);
396 /* Update potential sum for this i atom from the interaction with this j atom. */
397 velec = _mm_and_ps(velec,cutoff_mask);
398 velecsum = _mm_add_ps(velecsum,velec);
402 fscal = _mm_and_ps(fscal,cutoff_mask);
404 /* Calculate temporary vectorial force */
405 tx = _mm_mul_ps(fscal,dx01);
406 ty = _mm_mul_ps(fscal,dy01);
407 tz = _mm_mul_ps(fscal,dz01);
409 /* Update vectorial force */
410 fix0 = _mm_add_ps(fix0,tx);
411 fiy0 = _mm_add_ps(fiy0,ty);
412 fiz0 = _mm_add_ps(fiz0,tz);
414 fjx1 = _mm_add_ps(fjx1,tx);
415 fjy1 = _mm_add_ps(fjy1,ty);
416 fjz1 = _mm_add_ps(fjz1,tz);
420 /**************************
421 * CALCULATE INTERACTIONS *
422 **************************/
424 if (gmx_mm_any_lt(rsq02,rcutoff2))
427 r02 = _mm_mul_ps(rsq02,rinv02);
429 /* EWALD ELECTROSTATICS */
431 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
432 ewrt = _mm_mul_ps(r02,ewtabscale);
433 ewitab = _mm_cvttps_epi32(ewrt);
434 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
435 ewitab = _mm_slli_epi32(ewitab,2);
436 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
437 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
438 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
439 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
440 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
441 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
442 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
443 velec = _mm_mul_ps(qq02,_mm_sub_ps(rinv02,velec));
444 felec = _mm_mul_ps(_mm_mul_ps(qq02,rinv02),_mm_sub_ps(rinvsq02,felec));
446 d = _mm_sub_ps(r02,rswitch);
447 d = _mm_max_ps(d,_mm_setzero_ps());
448 d2 = _mm_mul_ps(d,d);
449 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
451 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
453 /* Evaluate switch function */
454 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
455 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv02,_mm_mul_ps(velec,dsw)) );
456 velec = _mm_mul_ps(velec,sw);
457 cutoff_mask = _mm_cmplt_ps(rsq02,rcutoff2);
459 /* Update potential sum for this i atom from the interaction with this j atom. */
460 velec = _mm_and_ps(velec,cutoff_mask);
461 velecsum = _mm_add_ps(velecsum,velec);
465 fscal = _mm_and_ps(fscal,cutoff_mask);
467 /* Calculate temporary vectorial force */
468 tx = _mm_mul_ps(fscal,dx02);
469 ty = _mm_mul_ps(fscal,dy02);
470 tz = _mm_mul_ps(fscal,dz02);
472 /* Update vectorial force */
473 fix0 = _mm_add_ps(fix0,tx);
474 fiy0 = _mm_add_ps(fiy0,ty);
475 fiz0 = _mm_add_ps(fiz0,tz);
477 fjx2 = _mm_add_ps(fjx2,tx);
478 fjy2 = _mm_add_ps(fjy2,ty);
479 fjz2 = _mm_add_ps(fjz2,tz);
483 /**************************
484 * CALCULATE INTERACTIONS *
485 **************************/
487 if (gmx_mm_any_lt(rsq10,rcutoff2))
490 r10 = _mm_mul_ps(rsq10,rinv10);
492 /* EWALD ELECTROSTATICS */
494 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
495 ewrt = _mm_mul_ps(r10,ewtabscale);
496 ewitab = _mm_cvttps_epi32(ewrt);
497 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
498 ewitab = _mm_slli_epi32(ewitab,2);
499 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
500 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
501 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
502 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
503 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
504 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
505 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
506 velec = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
507 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
509 d = _mm_sub_ps(r10,rswitch);
510 d = _mm_max_ps(d,_mm_setzero_ps());
511 d2 = _mm_mul_ps(d,d);
512 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
514 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
516 /* Evaluate switch function */
517 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
518 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
519 velec = _mm_mul_ps(velec,sw);
520 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
522 /* Update potential sum for this i atom from the interaction with this j atom. */
523 velec = _mm_and_ps(velec,cutoff_mask);
524 velecsum = _mm_add_ps(velecsum,velec);
528 fscal = _mm_and_ps(fscal,cutoff_mask);
530 /* Calculate temporary vectorial force */
531 tx = _mm_mul_ps(fscal,dx10);
532 ty = _mm_mul_ps(fscal,dy10);
533 tz = _mm_mul_ps(fscal,dz10);
535 /* Update vectorial force */
536 fix1 = _mm_add_ps(fix1,tx);
537 fiy1 = _mm_add_ps(fiy1,ty);
538 fiz1 = _mm_add_ps(fiz1,tz);
540 fjx0 = _mm_add_ps(fjx0,tx);
541 fjy0 = _mm_add_ps(fjy0,ty);
542 fjz0 = _mm_add_ps(fjz0,tz);
546 /**************************
547 * CALCULATE INTERACTIONS *
548 **************************/
550 if (gmx_mm_any_lt(rsq11,rcutoff2))
553 r11 = _mm_mul_ps(rsq11,rinv11);
555 /* EWALD ELECTROSTATICS */
557 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
558 ewrt = _mm_mul_ps(r11,ewtabscale);
559 ewitab = _mm_cvttps_epi32(ewrt);
560 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
561 ewitab = _mm_slli_epi32(ewitab,2);
562 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
563 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
564 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
565 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
566 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
567 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
568 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
569 velec = _mm_mul_ps(qq11,_mm_sub_ps(rinv11,velec));
570 felec = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
572 d = _mm_sub_ps(r11,rswitch);
573 d = _mm_max_ps(d,_mm_setzero_ps());
574 d2 = _mm_mul_ps(d,d);
575 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
577 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
579 /* Evaluate switch function */
580 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
581 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv11,_mm_mul_ps(velec,dsw)) );
582 velec = _mm_mul_ps(velec,sw);
583 cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2);
585 /* Update potential sum for this i atom from the interaction with this j atom. */
586 velec = _mm_and_ps(velec,cutoff_mask);
587 velecsum = _mm_add_ps(velecsum,velec);
591 fscal = _mm_and_ps(fscal,cutoff_mask);
593 /* Calculate temporary vectorial force */
594 tx = _mm_mul_ps(fscal,dx11);
595 ty = _mm_mul_ps(fscal,dy11);
596 tz = _mm_mul_ps(fscal,dz11);
598 /* Update vectorial force */
599 fix1 = _mm_add_ps(fix1,tx);
600 fiy1 = _mm_add_ps(fiy1,ty);
601 fiz1 = _mm_add_ps(fiz1,tz);
603 fjx1 = _mm_add_ps(fjx1,tx);
604 fjy1 = _mm_add_ps(fjy1,ty);
605 fjz1 = _mm_add_ps(fjz1,tz);
609 /**************************
610 * CALCULATE INTERACTIONS *
611 **************************/
613 if (gmx_mm_any_lt(rsq12,rcutoff2))
616 r12 = _mm_mul_ps(rsq12,rinv12);
618 /* EWALD ELECTROSTATICS */
620 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
621 ewrt = _mm_mul_ps(r12,ewtabscale);
622 ewitab = _mm_cvttps_epi32(ewrt);
623 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
624 ewitab = _mm_slli_epi32(ewitab,2);
625 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
626 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
627 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
628 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
629 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
630 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
631 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
632 velec = _mm_mul_ps(qq12,_mm_sub_ps(rinv12,velec));
633 felec = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
635 d = _mm_sub_ps(r12,rswitch);
636 d = _mm_max_ps(d,_mm_setzero_ps());
637 d2 = _mm_mul_ps(d,d);
638 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
640 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
642 /* Evaluate switch function */
643 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
644 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv12,_mm_mul_ps(velec,dsw)) );
645 velec = _mm_mul_ps(velec,sw);
646 cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2);
648 /* Update potential sum for this i atom from the interaction with this j atom. */
649 velec = _mm_and_ps(velec,cutoff_mask);
650 velecsum = _mm_add_ps(velecsum,velec);
654 fscal = _mm_and_ps(fscal,cutoff_mask);
656 /* Calculate temporary vectorial force */
657 tx = _mm_mul_ps(fscal,dx12);
658 ty = _mm_mul_ps(fscal,dy12);
659 tz = _mm_mul_ps(fscal,dz12);
661 /* Update vectorial force */
662 fix1 = _mm_add_ps(fix1,tx);
663 fiy1 = _mm_add_ps(fiy1,ty);
664 fiz1 = _mm_add_ps(fiz1,tz);
666 fjx2 = _mm_add_ps(fjx2,tx);
667 fjy2 = _mm_add_ps(fjy2,ty);
668 fjz2 = _mm_add_ps(fjz2,tz);
672 /**************************
673 * CALCULATE INTERACTIONS *
674 **************************/
676 if (gmx_mm_any_lt(rsq20,rcutoff2))
679 r20 = _mm_mul_ps(rsq20,rinv20);
681 /* EWALD ELECTROSTATICS */
683 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
684 ewrt = _mm_mul_ps(r20,ewtabscale);
685 ewitab = _mm_cvttps_epi32(ewrt);
686 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
687 ewitab = _mm_slli_epi32(ewitab,2);
688 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
689 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
690 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
691 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
692 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
693 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
694 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
695 velec = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
696 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
698 d = _mm_sub_ps(r20,rswitch);
699 d = _mm_max_ps(d,_mm_setzero_ps());
700 d2 = _mm_mul_ps(d,d);
701 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
703 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
705 /* Evaluate switch function */
706 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
707 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
708 velec = _mm_mul_ps(velec,sw);
709 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
711 /* Update potential sum for this i atom from the interaction with this j atom. */
712 velec = _mm_and_ps(velec,cutoff_mask);
713 velecsum = _mm_add_ps(velecsum,velec);
717 fscal = _mm_and_ps(fscal,cutoff_mask);
719 /* Calculate temporary vectorial force */
720 tx = _mm_mul_ps(fscal,dx20);
721 ty = _mm_mul_ps(fscal,dy20);
722 tz = _mm_mul_ps(fscal,dz20);
724 /* Update vectorial force */
725 fix2 = _mm_add_ps(fix2,tx);
726 fiy2 = _mm_add_ps(fiy2,ty);
727 fiz2 = _mm_add_ps(fiz2,tz);
729 fjx0 = _mm_add_ps(fjx0,tx);
730 fjy0 = _mm_add_ps(fjy0,ty);
731 fjz0 = _mm_add_ps(fjz0,tz);
735 /**************************
736 * CALCULATE INTERACTIONS *
737 **************************/
739 if (gmx_mm_any_lt(rsq21,rcutoff2))
742 r21 = _mm_mul_ps(rsq21,rinv21);
744 /* EWALD ELECTROSTATICS */
746 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
747 ewrt = _mm_mul_ps(r21,ewtabscale);
748 ewitab = _mm_cvttps_epi32(ewrt);
749 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
750 ewitab = _mm_slli_epi32(ewitab,2);
751 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
752 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
753 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
754 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
755 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
756 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
757 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
758 velec = _mm_mul_ps(qq21,_mm_sub_ps(rinv21,velec));
759 felec = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
761 d = _mm_sub_ps(r21,rswitch);
762 d = _mm_max_ps(d,_mm_setzero_ps());
763 d2 = _mm_mul_ps(d,d);
764 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
766 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
768 /* Evaluate switch function */
769 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
770 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv21,_mm_mul_ps(velec,dsw)) );
771 velec = _mm_mul_ps(velec,sw);
772 cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2);
774 /* Update potential sum for this i atom from the interaction with this j atom. */
775 velec = _mm_and_ps(velec,cutoff_mask);
776 velecsum = _mm_add_ps(velecsum,velec);
780 fscal = _mm_and_ps(fscal,cutoff_mask);
782 /* Calculate temporary vectorial force */
783 tx = _mm_mul_ps(fscal,dx21);
784 ty = _mm_mul_ps(fscal,dy21);
785 tz = _mm_mul_ps(fscal,dz21);
787 /* Update vectorial force */
788 fix2 = _mm_add_ps(fix2,tx);
789 fiy2 = _mm_add_ps(fiy2,ty);
790 fiz2 = _mm_add_ps(fiz2,tz);
792 fjx1 = _mm_add_ps(fjx1,tx);
793 fjy1 = _mm_add_ps(fjy1,ty);
794 fjz1 = _mm_add_ps(fjz1,tz);
798 /**************************
799 * CALCULATE INTERACTIONS *
800 **************************/
802 if (gmx_mm_any_lt(rsq22,rcutoff2))
805 r22 = _mm_mul_ps(rsq22,rinv22);
807 /* EWALD ELECTROSTATICS */
809 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
810 ewrt = _mm_mul_ps(r22,ewtabscale);
811 ewitab = _mm_cvttps_epi32(ewrt);
812 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
813 ewitab = _mm_slli_epi32(ewitab,2);
814 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
815 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
816 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
817 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
818 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
819 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
820 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
821 velec = _mm_mul_ps(qq22,_mm_sub_ps(rinv22,velec));
822 felec = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
824 d = _mm_sub_ps(r22,rswitch);
825 d = _mm_max_ps(d,_mm_setzero_ps());
826 d2 = _mm_mul_ps(d,d);
827 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
829 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
831 /* Evaluate switch function */
832 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
833 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv22,_mm_mul_ps(velec,dsw)) );
834 velec = _mm_mul_ps(velec,sw);
835 cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2);
837 /* Update potential sum for this i atom from the interaction with this j atom. */
838 velec = _mm_and_ps(velec,cutoff_mask);
839 velecsum = _mm_add_ps(velecsum,velec);
843 fscal = _mm_and_ps(fscal,cutoff_mask);
845 /* Calculate temporary vectorial force */
846 tx = _mm_mul_ps(fscal,dx22);
847 ty = _mm_mul_ps(fscal,dy22);
848 tz = _mm_mul_ps(fscal,dz22);
850 /* Update vectorial force */
851 fix2 = _mm_add_ps(fix2,tx);
852 fiy2 = _mm_add_ps(fiy2,ty);
853 fiz2 = _mm_add_ps(fiz2,tz);
855 fjx2 = _mm_add_ps(fjx2,tx);
856 fjy2 = _mm_add_ps(fjy2,ty);
857 fjz2 = _mm_add_ps(fjz2,tz);
861 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
862 f+j_coord_offsetC,f+j_coord_offsetD,
863 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
865 /* Inner loop uses 585 flops */
871 /* Get j neighbor index, and coordinate index */
877 /* Sign of each element will be negative for non-real atoms.
878 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
879 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
881 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
882 jnrA = (jnrA>=0) ? jnrA : 0;
883 jnrB = (jnrB>=0) ? jnrB : 0;
884 jnrC = (jnrC>=0) ? jnrC : 0;
885 jnrD = (jnrD>=0) ? jnrD : 0;
887 j_coord_offsetA = DIM*jnrA;
888 j_coord_offsetB = DIM*jnrB;
889 j_coord_offsetC = DIM*jnrC;
890 j_coord_offsetD = DIM*jnrD;
892 /* load j atom coordinates */
893 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
894 x+j_coord_offsetC,x+j_coord_offsetD,
895 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
897 /* Calculate displacement vector */
898 dx00 = _mm_sub_ps(ix0,jx0);
899 dy00 = _mm_sub_ps(iy0,jy0);
900 dz00 = _mm_sub_ps(iz0,jz0);
901 dx01 = _mm_sub_ps(ix0,jx1);
902 dy01 = _mm_sub_ps(iy0,jy1);
903 dz01 = _mm_sub_ps(iz0,jz1);
904 dx02 = _mm_sub_ps(ix0,jx2);
905 dy02 = _mm_sub_ps(iy0,jy2);
906 dz02 = _mm_sub_ps(iz0,jz2);
907 dx10 = _mm_sub_ps(ix1,jx0);
908 dy10 = _mm_sub_ps(iy1,jy0);
909 dz10 = _mm_sub_ps(iz1,jz0);
910 dx11 = _mm_sub_ps(ix1,jx1);
911 dy11 = _mm_sub_ps(iy1,jy1);
912 dz11 = _mm_sub_ps(iz1,jz1);
913 dx12 = _mm_sub_ps(ix1,jx2);
914 dy12 = _mm_sub_ps(iy1,jy2);
915 dz12 = _mm_sub_ps(iz1,jz2);
916 dx20 = _mm_sub_ps(ix2,jx0);
917 dy20 = _mm_sub_ps(iy2,jy0);
918 dz20 = _mm_sub_ps(iz2,jz0);
919 dx21 = _mm_sub_ps(ix2,jx1);
920 dy21 = _mm_sub_ps(iy2,jy1);
921 dz21 = _mm_sub_ps(iz2,jz1);
922 dx22 = _mm_sub_ps(ix2,jx2);
923 dy22 = _mm_sub_ps(iy2,jy2);
924 dz22 = _mm_sub_ps(iz2,jz2);
926 /* Calculate squared distance and things based on it */
927 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
928 rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
929 rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
930 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
931 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
932 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
933 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
934 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
935 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
937 rinv00 = gmx_mm_invsqrt_ps(rsq00);
938 rinv01 = gmx_mm_invsqrt_ps(rsq01);
939 rinv02 = gmx_mm_invsqrt_ps(rsq02);
940 rinv10 = gmx_mm_invsqrt_ps(rsq10);
941 rinv11 = gmx_mm_invsqrt_ps(rsq11);
942 rinv12 = gmx_mm_invsqrt_ps(rsq12);
943 rinv20 = gmx_mm_invsqrt_ps(rsq20);
944 rinv21 = gmx_mm_invsqrt_ps(rsq21);
945 rinv22 = gmx_mm_invsqrt_ps(rsq22);
947 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
948 rinvsq01 = _mm_mul_ps(rinv01,rinv01);
949 rinvsq02 = _mm_mul_ps(rinv02,rinv02);
950 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
951 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
952 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
953 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
954 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
955 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
957 fjx0 = _mm_setzero_ps();
958 fjy0 = _mm_setzero_ps();
959 fjz0 = _mm_setzero_ps();
960 fjx1 = _mm_setzero_ps();
961 fjy1 = _mm_setzero_ps();
962 fjz1 = _mm_setzero_ps();
963 fjx2 = _mm_setzero_ps();
964 fjy2 = _mm_setzero_ps();
965 fjz2 = _mm_setzero_ps();
967 /**************************
968 * CALCULATE INTERACTIONS *
969 **************************/
971 if (gmx_mm_any_lt(rsq00,rcutoff2))
974 r00 = _mm_mul_ps(rsq00,rinv00);
975 r00 = _mm_andnot_ps(dummy_mask,r00);
977 /* EWALD ELECTROSTATICS */
979 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
980 ewrt = _mm_mul_ps(r00,ewtabscale);
981 ewitab = _mm_cvttps_epi32(ewrt);
982 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
983 ewitab = _mm_slli_epi32(ewitab,2);
984 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
985 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
986 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
987 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
988 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
989 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
990 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
991 velec = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
992 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
994 d = _mm_sub_ps(r00,rswitch);
995 d = _mm_max_ps(d,_mm_setzero_ps());
996 d2 = _mm_mul_ps(d,d);
997 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
999 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1001 /* Evaluate switch function */
1002 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1003 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1004 velec = _mm_mul_ps(velec,sw);
1005 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1007 /* Update potential sum for this i atom from the interaction with this j atom. */
1008 velec = _mm_and_ps(velec,cutoff_mask);
1009 velec = _mm_andnot_ps(dummy_mask,velec);
1010 velecsum = _mm_add_ps(velecsum,velec);
1014 fscal = _mm_and_ps(fscal,cutoff_mask);
1016 fscal = _mm_andnot_ps(dummy_mask,fscal);
1018 /* Calculate temporary vectorial force */
1019 tx = _mm_mul_ps(fscal,dx00);
1020 ty = _mm_mul_ps(fscal,dy00);
1021 tz = _mm_mul_ps(fscal,dz00);
1023 /* Update vectorial force */
1024 fix0 = _mm_add_ps(fix0,tx);
1025 fiy0 = _mm_add_ps(fiy0,ty);
1026 fiz0 = _mm_add_ps(fiz0,tz);
1028 fjx0 = _mm_add_ps(fjx0,tx);
1029 fjy0 = _mm_add_ps(fjy0,ty);
1030 fjz0 = _mm_add_ps(fjz0,tz);
1034 /**************************
1035 * CALCULATE INTERACTIONS *
1036 **************************/
1038 if (gmx_mm_any_lt(rsq01,rcutoff2))
1041 r01 = _mm_mul_ps(rsq01,rinv01);
1042 r01 = _mm_andnot_ps(dummy_mask,r01);
1044 /* EWALD ELECTROSTATICS */
1046 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1047 ewrt = _mm_mul_ps(r01,ewtabscale);
1048 ewitab = _mm_cvttps_epi32(ewrt);
1049 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1050 ewitab = _mm_slli_epi32(ewitab,2);
1051 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1052 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1053 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1054 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1055 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1056 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1057 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1058 velec = _mm_mul_ps(qq01,_mm_sub_ps(rinv01,velec));
1059 felec = _mm_mul_ps(_mm_mul_ps(qq01,rinv01),_mm_sub_ps(rinvsq01,felec));
1061 d = _mm_sub_ps(r01,rswitch);
1062 d = _mm_max_ps(d,_mm_setzero_ps());
1063 d2 = _mm_mul_ps(d,d);
1064 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1066 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1068 /* Evaluate switch function */
1069 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1070 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv01,_mm_mul_ps(velec,dsw)) );
1071 velec = _mm_mul_ps(velec,sw);
1072 cutoff_mask = _mm_cmplt_ps(rsq01,rcutoff2);
1074 /* Update potential sum for this i atom from the interaction with this j atom. */
1075 velec = _mm_and_ps(velec,cutoff_mask);
1076 velec = _mm_andnot_ps(dummy_mask,velec);
1077 velecsum = _mm_add_ps(velecsum,velec);
1081 fscal = _mm_and_ps(fscal,cutoff_mask);
1083 fscal = _mm_andnot_ps(dummy_mask,fscal);
1085 /* Calculate temporary vectorial force */
1086 tx = _mm_mul_ps(fscal,dx01);
1087 ty = _mm_mul_ps(fscal,dy01);
1088 tz = _mm_mul_ps(fscal,dz01);
1090 /* Update vectorial force */
1091 fix0 = _mm_add_ps(fix0,tx);
1092 fiy0 = _mm_add_ps(fiy0,ty);
1093 fiz0 = _mm_add_ps(fiz0,tz);
1095 fjx1 = _mm_add_ps(fjx1,tx);
1096 fjy1 = _mm_add_ps(fjy1,ty);
1097 fjz1 = _mm_add_ps(fjz1,tz);
1101 /**************************
1102 * CALCULATE INTERACTIONS *
1103 **************************/
1105 if (gmx_mm_any_lt(rsq02,rcutoff2))
1108 r02 = _mm_mul_ps(rsq02,rinv02);
1109 r02 = _mm_andnot_ps(dummy_mask,r02);
1111 /* EWALD ELECTROSTATICS */
1113 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1114 ewrt = _mm_mul_ps(r02,ewtabscale);
1115 ewitab = _mm_cvttps_epi32(ewrt);
1116 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1117 ewitab = _mm_slli_epi32(ewitab,2);
1118 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1119 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1120 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1121 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1122 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1123 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1124 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1125 velec = _mm_mul_ps(qq02,_mm_sub_ps(rinv02,velec));
1126 felec = _mm_mul_ps(_mm_mul_ps(qq02,rinv02),_mm_sub_ps(rinvsq02,felec));
1128 d = _mm_sub_ps(r02,rswitch);
1129 d = _mm_max_ps(d,_mm_setzero_ps());
1130 d2 = _mm_mul_ps(d,d);
1131 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1133 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1135 /* Evaluate switch function */
1136 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1137 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv02,_mm_mul_ps(velec,dsw)) );
1138 velec = _mm_mul_ps(velec,sw);
1139 cutoff_mask = _mm_cmplt_ps(rsq02,rcutoff2);
1141 /* Update potential sum for this i atom from the interaction with this j atom. */
1142 velec = _mm_and_ps(velec,cutoff_mask);
1143 velec = _mm_andnot_ps(dummy_mask,velec);
1144 velecsum = _mm_add_ps(velecsum,velec);
1148 fscal = _mm_and_ps(fscal,cutoff_mask);
1150 fscal = _mm_andnot_ps(dummy_mask,fscal);
1152 /* Calculate temporary vectorial force */
1153 tx = _mm_mul_ps(fscal,dx02);
1154 ty = _mm_mul_ps(fscal,dy02);
1155 tz = _mm_mul_ps(fscal,dz02);
1157 /* Update vectorial force */
1158 fix0 = _mm_add_ps(fix0,tx);
1159 fiy0 = _mm_add_ps(fiy0,ty);
1160 fiz0 = _mm_add_ps(fiz0,tz);
1162 fjx2 = _mm_add_ps(fjx2,tx);
1163 fjy2 = _mm_add_ps(fjy2,ty);
1164 fjz2 = _mm_add_ps(fjz2,tz);
1168 /**************************
1169 * CALCULATE INTERACTIONS *
1170 **************************/
1172 if (gmx_mm_any_lt(rsq10,rcutoff2))
1175 r10 = _mm_mul_ps(rsq10,rinv10);
1176 r10 = _mm_andnot_ps(dummy_mask,r10);
1178 /* EWALD ELECTROSTATICS */
1180 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1181 ewrt = _mm_mul_ps(r10,ewtabscale);
1182 ewitab = _mm_cvttps_epi32(ewrt);
1183 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1184 ewitab = _mm_slli_epi32(ewitab,2);
1185 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1186 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1187 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1188 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1189 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1190 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1191 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1192 velec = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1193 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1195 d = _mm_sub_ps(r10,rswitch);
1196 d = _mm_max_ps(d,_mm_setzero_ps());
1197 d2 = _mm_mul_ps(d,d);
1198 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1200 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1202 /* Evaluate switch function */
1203 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1204 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1205 velec = _mm_mul_ps(velec,sw);
1206 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
1208 /* Update potential sum for this i atom from the interaction with this j atom. */
1209 velec = _mm_and_ps(velec,cutoff_mask);
1210 velec = _mm_andnot_ps(dummy_mask,velec);
1211 velecsum = _mm_add_ps(velecsum,velec);
1215 fscal = _mm_and_ps(fscal,cutoff_mask);
1217 fscal = _mm_andnot_ps(dummy_mask,fscal);
1219 /* Calculate temporary vectorial force */
1220 tx = _mm_mul_ps(fscal,dx10);
1221 ty = _mm_mul_ps(fscal,dy10);
1222 tz = _mm_mul_ps(fscal,dz10);
1224 /* Update vectorial force */
1225 fix1 = _mm_add_ps(fix1,tx);
1226 fiy1 = _mm_add_ps(fiy1,ty);
1227 fiz1 = _mm_add_ps(fiz1,tz);
1229 fjx0 = _mm_add_ps(fjx0,tx);
1230 fjy0 = _mm_add_ps(fjy0,ty);
1231 fjz0 = _mm_add_ps(fjz0,tz);
1235 /**************************
1236 * CALCULATE INTERACTIONS *
1237 **************************/
1239 if (gmx_mm_any_lt(rsq11,rcutoff2))
1242 r11 = _mm_mul_ps(rsq11,rinv11);
1243 r11 = _mm_andnot_ps(dummy_mask,r11);
1245 /* EWALD ELECTROSTATICS */
1247 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1248 ewrt = _mm_mul_ps(r11,ewtabscale);
1249 ewitab = _mm_cvttps_epi32(ewrt);
1250 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1251 ewitab = _mm_slli_epi32(ewitab,2);
1252 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1253 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1254 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1255 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1256 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1257 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1258 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1259 velec = _mm_mul_ps(qq11,_mm_sub_ps(rinv11,velec));
1260 felec = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
1262 d = _mm_sub_ps(r11,rswitch);
1263 d = _mm_max_ps(d,_mm_setzero_ps());
1264 d2 = _mm_mul_ps(d,d);
1265 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1267 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1269 /* Evaluate switch function */
1270 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1271 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv11,_mm_mul_ps(velec,dsw)) );
1272 velec = _mm_mul_ps(velec,sw);
1273 cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2);
1275 /* Update potential sum for this i atom from the interaction with this j atom. */
1276 velec = _mm_and_ps(velec,cutoff_mask);
1277 velec = _mm_andnot_ps(dummy_mask,velec);
1278 velecsum = _mm_add_ps(velecsum,velec);
1282 fscal = _mm_and_ps(fscal,cutoff_mask);
1284 fscal = _mm_andnot_ps(dummy_mask,fscal);
1286 /* Calculate temporary vectorial force */
1287 tx = _mm_mul_ps(fscal,dx11);
1288 ty = _mm_mul_ps(fscal,dy11);
1289 tz = _mm_mul_ps(fscal,dz11);
1291 /* Update vectorial force */
1292 fix1 = _mm_add_ps(fix1,tx);
1293 fiy1 = _mm_add_ps(fiy1,ty);
1294 fiz1 = _mm_add_ps(fiz1,tz);
1296 fjx1 = _mm_add_ps(fjx1,tx);
1297 fjy1 = _mm_add_ps(fjy1,ty);
1298 fjz1 = _mm_add_ps(fjz1,tz);
1302 /**************************
1303 * CALCULATE INTERACTIONS *
1304 **************************/
1306 if (gmx_mm_any_lt(rsq12,rcutoff2))
1309 r12 = _mm_mul_ps(rsq12,rinv12);
1310 r12 = _mm_andnot_ps(dummy_mask,r12);
1312 /* EWALD ELECTROSTATICS */
1314 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1315 ewrt = _mm_mul_ps(r12,ewtabscale);
1316 ewitab = _mm_cvttps_epi32(ewrt);
1317 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1318 ewitab = _mm_slli_epi32(ewitab,2);
1319 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1320 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1321 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1322 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1323 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1324 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1325 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1326 velec = _mm_mul_ps(qq12,_mm_sub_ps(rinv12,velec));
1327 felec = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
1329 d = _mm_sub_ps(r12,rswitch);
1330 d = _mm_max_ps(d,_mm_setzero_ps());
1331 d2 = _mm_mul_ps(d,d);
1332 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1334 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1336 /* Evaluate switch function */
1337 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1338 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv12,_mm_mul_ps(velec,dsw)) );
1339 velec = _mm_mul_ps(velec,sw);
1340 cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2);
1342 /* Update potential sum for this i atom from the interaction with this j atom. */
1343 velec = _mm_and_ps(velec,cutoff_mask);
1344 velec = _mm_andnot_ps(dummy_mask,velec);
1345 velecsum = _mm_add_ps(velecsum,velec);
1349 fscal = _mm_and_ps(fscal,cutoff_mask);
1351 fscal = _mm_andnot_ps(dummy_mask,fscal);
1353 /* Calculate temporary vectorial force */
1354 tx = _mm_mul_ps(fscal,dx12);
1355 ty = _mm_mul_ps(fscal,dy12);
1356 tz = _mm_mul_ps(fscal,dz12);
1358 /* Update vectorial force */
1359 fix1 = _mm_add_ps(fix1,tx);
1360 fiy1 = _mm_add_ps(fiy1,ty);
1361 fiz1 = _mm_add_ps(fiz1,tz);
1363 fjx2 = _mm_add_ps(fjx2,tx);
1364 fjy2 = _mm_add_ps(fjy2,ty);
1365 fjz2 = _mm_add_ps(fjz2,tz);
1369 /**************************
1370 * CALCULATE INTERACTIONS *
1371 **************************/
1373 if (gmx_mm_any_lt(rsq20,rcutoff2))
1376 r20 = _mm_mul_ps(rsq20,rinv20);
1377 r20 = _mm_andnot_ps(dummy_mask,r20);
1379 /* EWALD ELECTROSTATICS */
1381 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1382 ewrt = _mm_mul_ps(r20,ewtabscale);
1383 ewitab = _mm_cvttps_epi32(ewrt);
1384 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1385 ewitab = _mm_slli_epi32(ewitab,2);
1386 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1387 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1388 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1389 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1390 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1391 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1392 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1393 velec = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1394 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1396 d = _mm_sub_ps(r20,rswitch);
1397 d = _mm_max_ps(d,_mm_setzero_ps());
1398 d2 = _mm_mul_ps(d,d);
1399 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1401 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1403 /* Evaluate switch function */
1404 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1405 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1406 velec = _mm_mul_ps(velec,sw);
1407 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
1409 /* Update potential sum for this i atom from the interaction with this j atom. */
1410 velec = _mm_and_ps(velec,cutoff_mask);
1411 velec = _mm_andnot_ps(dummy_mask,velec);
1412 velecsum = _mm_add_ps(velecsum,velec);
1416 fscal = _mm_and_ps(fscal,cutoff_mask);
1418 fscal = _mm_andnot_ps(dummy_mask,fscal);
1420 /* Calculate temporary vectorial force */
1421 tx = _mm_mul_ps(fscal,dx20);
1422 ty = _mm_mul_ps(fscal,dy20);
1423 tz = _mm_mul_ps(fscal,dz20);
1425 /* Update vectorial force */
1426 fix2 = _mm_add_ps(fix2,tx);
1427 fiy2 = _mm_add_ps(fiy2,ty);
1428 fiz2 = _mm_add_ps(fiz2,tz);
1430 fjx0 = _mm_add_ps(fjx0,tx);
1431 fjy0 = _mm_add_ps(fjy0,ty);
1432 fjz0 = _mm_add_ps(fjz0,tz);
1436 /**************************
1437 * CALCULATE INTERACTIONS *
1438 **************************/
1440 if (gmx_mm_any_lt(rsq21,rcutoff2))
1443 r21 = _mm_mul_ps(rsq21,rinv21);
1444 r21 = _mm_andnot_ps(dummy_mask,r21);
1446 /* EWALD ELECTROSTATICS */
1448 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1449 ewrt = _mm_mul_ps(r21,ewtabscale);
1450 ewitab = _mm_cvttps_epi32(ewrt);
1451 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1452 ewitab = _mm_slli_epi32(ewitab,2);
1453 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1454 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1455 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1456 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1457 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1458 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1459 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1460 velec = _mm_mul_ps(qq21,_mm_sub_ps(rinv21,velec));
1461 felec = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
1463 d = _mm_sub_ps(r21,rswitch);
1464 d = _mm_max_ps(d,_mm_setzero_ps());
1465 d2 = _mm_mul_ps(d,d);
1466 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1468 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1470 /* Evaluate switch function */
1471 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1472 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv21,_mm_mul_ps(velec,dsw)) );
1473 velec = _mm_mul_ps(velec,sw);
1474 cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2);
1476 /* Update potential sum for this i atom from the interaction with this j atom. */
1477 velec = _mm_and_ps(velec,cutoff_mask);
1478 velec = _mm_andnot_ps(dummy_mask,velec);
1479 velecsum = _mm_add_ps(velecsum,velec);
1483 fscal = _mm_and_ps(fscal,cutoff_mask);
1485 fscal = _mm_andnot_ps(dummy_mask,fscal);
1487 /* Calculate temporary vectorial force */
1488 tx = _mm_mul_ps(fscal,dx21);
1489 ty = _mm_mul_ps(fscal,dy21);
1490 tz = _mm_mul_ps(fscal,dz21);
1492 /* Update vectorial force */
1493 fix2 = _mm_add_ps(fix2,tx);
1494 fiy2 = _mm_add_ps(fiy2,ty);
1495 fiz2 = _mm_add_ps(fiz2,tz);
1497 fjx1 = _mm_add_ps(fjx1,tx);
1498 fjy1 = _mm_add_ps(fjy1,ty);
1499 fjz1 = _mm_add_ps(fjz1,tz);
1503 /**************************
1504 * CALCULATE INTERACTIONS *
1505 **************************/
1507 if (gmx_mm_any_lt(rsq22,rcutoff2))
1510 r22 = _mm_mul_ps(rsq22,rinv22);
1511 r22 = _mm_andnot_ps(dummy_mask,r22);
1513 /* EWALD ELECTROSTATICS */
1515 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1516 ewrt = _mm_mul_ps(r22,ewtabscale);
1517 ewitab = _mm_cvttps_epi32(ewrt);
1518 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1519 ewitab = _mm_slli_epi32(ewitab,2);
1520 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1521 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1522 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1523 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1524 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1525 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1526 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1527 velec = _mm_mul_ps(qq22,_mm_sub_ps(rinv22,velec));
1528 felec = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
1530 d = _mm_sub_ps(r22,rswitch);
1531 d = _mm_max_ps(d,_mm_setzero_ps());
1532 d2 = _mm_mul_ps(d,d);
1533 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1535 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1537 /* Evaluate switch function */
1538 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1539 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv22,_mm_mul_ps(velec,dsw)) );
1540 velec = _mm_mul_ps(velec,sw);
1541 cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2);
1543 /* Update potential sum for this i atom from the interaction with this j atom. */
1544 velec = _mm_and_ps(velec,cutoff_mask);
1545 velec = _mm_andnot_ps(dummy_mask,velec);
1546 velecsum = _mm_add_ps(velecsum,velec);
1550 fscal = _mm_and_ps(fscal,cutoff_mask);
1552 fscal = _mm_andnot_ps(dummy_mask,fscal);
1554 /* Calculate temporary vectorial force */
1555 tx = _mm_mul_ps(fscal,dx22);
1556 ty = _mm_mul_ps(fscal,dy22);
1557 tz = _mm_mul_ps(fscal,dz22);
1559 /* Update vectorial force */
1560 fix2 = _mm_add_ps(fix2,tx);
1561 fiy2 = _mm_add_ps(fiy2,ty);
1562 fiz2 = _mm_add_ps(fiz2,tz);
1564 fjx2 = _mm_add_ps(fjx2,tx);
1565 fjy2 = _mm_add_ps(fjy2,ty);
1566 fjz2 = _mm_add_ps(fjz2,tz);
1570 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1571 f+j_coord_offsetC,f+j_coord_offsetD,
1572 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1574 /* Inner loop uses 594 flops */
1577 /* End of innermost loop */
1579 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1580 f+i_coord_offset,fshift+i_shift_offset);
1583 /* Update potential energies */
1584 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1586 /* Increment number of inner iterations */
1587 inneriter += j_index_end - j_index_start;
1589 /* Outer loop uses 28 flops */
1592 /* Increment number of outer iterations */
1595 /* Update outer/inner flops */
1597 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*28 + inneriter*594);
1600 * Gromacs nonbonded kernel: nb_kernel_ElecEwSw_VdwNone_GeomW3W3_F_sse2_single
1601 * Electrostatics interaction: Ewald
1602 * VdW interaction: None
1603 * Geometry: Water3-Water3
1604 * Calculate force/pot: Force
1607 nb_kernel_ElecEwSw_VdwNone_GeomW3W3_F_sse2_single
1608 (t_nblist * gmx_restrict nlist,
1609 rvec * gmx_restrict xx,
1610 rvec * gmx_restrict ff,
1611 t_forcerec * gmx_restrict fr,
1612 t_mdatoms * gmx_restrict mdatoms,
1613 nb_kernel_data_t * gmx_restrict kernel_data,
1614 t_nrnb * gmx_restrict nrnb)
1616 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1617 * just 0 for non-waters.
1618 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
1619 * jnr indices corresponding to data put in the four positions in the SIMD register.
1621 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1622 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1623 int jnrA,jnrB,jnrC,jnrD;
1624 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1625 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1626 real shX,shY,shZ,rcutoff_scalar;
1627 real *shiftvec,*fshift,*x,*f;
1628 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1630 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1632 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1634 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1635 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1636 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1637 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1638 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1639 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1640 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1641 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1642 __m128 dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1643 __m128 dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1644 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1645 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1646 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1647 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1648 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1649 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1650 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
1653 __m128 ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1655 __m128 rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
1656 real rswitch_scalar,d_scalar;
1657 __m128 dummy_mask,cutoff_mask;
1658 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1659 __m128 one = _mm_set1_ps(1.0);
1660 __m128 two = _mm_set1_ps(2.0);
1666 jindex = nlist->jindex;
1668 shiftidx = nlist->shift;
1670 shiftvec = fr->shift_vec[0];
1671 fshift = fr->fshift[0];
1672 facel = _mm_set1_ps(fr->epsfac);
1673 charge = mdatoms->chargeA;
1675 sh_ewald = _mm_set1_ps(fr->ic->sh_ewald);
1676 ewtab = fr->ic->tabq_coul_FDV0;
1677 ewtabscale = _mm_set1_ps(fr->ic->tabq_scale);
1678 ewtabhalfspace = _mm_set1_ps(0.5/fr->ic->tabq_scale);
1680 /* Setup water-specific parameters */
1681 inr = nlist->iinr[0];
1682 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
1683 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1684 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1686 jq0 = _mm_set1_ps(charge[inr+0]);
1687 jq1 = _mm_set1_ps(charge[inr+1]);
1688 jq2 = _mm_set1_ps(charge[inr+2]);
1689 qq00 = _mm_mul_ps(iq0,jq0);
1690 qq01 = _mm_mul_ps(iq0,jq1);
1691 qq02 = _mm_mul_ps(iq0,jq2);
1692 qq10 = _mm_mul_ps(iq1,jq0);
1693 qq11 = _mm_mul_ps(iq1,jq1);
1694 qq12 = _mm_mul_ps(iq1,jq2);
1695 qq20 = _mm_mul_ps(iq2,jq0);
1696 qq21 = _mm_mul_ps(iq2,jq1);
1697 qq22 = _mm_mul_ps(iq2,jq2);
1699 /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1700 rcutoff_scalar = fr->rcoulomb;
1701 rcutoff = _mm_set1_ps(rcutoff_scalar);
1702 rcutoff2 = _mm_mul_ps(rcutoff,rcutoff);
1704 rswitch_scalar = fr->rcoulomb_switch;
1705 rswitch = _mm_set1_ps(rswitch_scalar);
1706 /* Setup switch parameters */
1707 d_scalar = rcutoff_scalar-rswitch_scalar;
1708 d = _mm_set1_ps(d_scalar);
1709 swV3 = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
1710 swV4 = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1711 swV5 = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1712 swF2 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
1713 swF3 = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
1714 swF4 = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
1716 /* Avoid stupid compiler warnings */
1717 jnrA = jnrB = jnrC = jnrD = 0;
1718 j_coord_offsetA = 0;
1719 j_coord_offsetB = 0;
1720 j_coord_offsetC = 0;
1721 j_coord_offsetD = 0;
1726 /* Start outer loop over neighborlists */
1727 for(iidx=0; iidx<nri; iidx++)
1729 /* Load shift vector for this list */
1730 i_shift_offset = DIM*shiftidx[iidx];
1731 shX = shiftvec[i_shift_offset+XX];
1732 shY = shiftvec[i_shift_offset+YY];
1733 shZ = shiftvec[i_shift_offset+ZZ];
1735 /* Load limits for loop over neighbors */
1736 j_index_start = jindex[iidx];
1737 j_index_end = jindex[iidx+1];
1739 /* Get outer coordinate index */
1741 i_coord_offset = DIM*inr;
1743 /* Load i particle coords and add shift vector */
1744 ix0 = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
1745 iy0 = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
1746 iz0 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
1747 ix1 = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
1748 iy1 = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
1749 iz1 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
1750 ix2 = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
1751 iy2 = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
1752 iz2 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
1754 fix0 = _mm_setzero_ps();
1755 fiy0 = _mm_setzero_ps();
1756 fiz0 = _mm_setzero_ps();
1757 fix1 = _mm_setzero_ps();
1758 fiy1 = _mm_setzero_ps();
1759 fiz1 = _mm_setzero_ps();
1760 fix2 = _mm_setzero_ps();
1761 fiy2 = _mm_setzero_ps();
1762 fiz2 = _mm_setzero_ps();
1764 /* Start inner kernel loop */
1765 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1768 /* Get j neighbor index, and coordinate index */
1770 jnrB = jjnr[jidx+1];
1771 jnrC = jjnr[jidx+2];
1772 jnrD = jjnr[jidx+3];
1774 j_coord_offsetA = DIM*jnrA;
1775 j_coord_offsetB = DIM*jnrB;
1776 j_coord_offsetC = DIM*jnrC;
1777 j_coord_offsetD = DIM*jnrD;
1779 /* load j atom coordinates */
1780 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1781 x+j_coord_offsetC,x+j_coord_offsetD,
1782 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1784 /* Calculate displacement vector */
1785 dx00 = _mm_sub_ps(ix0,jx0);
1786 dy00 = _mm_sub_ps(iy0,jy0);
1787 dz00 = _mm_sub_ps(iz0,jz0);
1788 dx01 = _mm_sub_ps(ix0,jx1);
1789 dy01 = _mm_sub_ps(iy0,jy1);
1790 dz01 = _mm_sub_ps(iz0,jz1);
1791 dx02 = _mm_sub_ps(ix0,jx2);
1792 dy02 = _mm_sub_ps(iy0,jy2);
1793 dz02 = _mm_sub_ps(iz0,jz2);
1794 dx10 = _mm_sub_ps(ix1,jx0);
1795 dy10 = _mm_sub_ps(iy1,jy0);
1796 dz10 = _mm_sub_ps(iz1,jz0);
1797 dx11 = _mm_sub_ps(ix1,jx1);
1798 dy11 = _mm_sub_ps(iy1,jy1);
1799 dz11 = _mm_sub_ps(iz1,jz1);
1800 dx12 = _mm_sub_ps(ix1,jx2);
1801 dy12 = _mm_sub_ps(iy1,jy2);
1802 dz12 = _mm_sub_ps(iz1,jz2);
1803 dx20 = _mm_sub_ps(ix2,jx0);
1804 dy20 = _mm_sub_ps(iy2,jy0);
1805 dz20 = _mm_sub_ps(iz2,jz0);
1806 dx21 = _mm_sub_ps(ix2,jx1);
1807 dy21 = _mm_sub_ps(iy2,jy1);
1808 dz21 = _mm_sub_ps(iz2,jz1);
1809 dx22 = _mm_sub_ps(ix2,jx2);
1810 dy22 = _mm_sub_ps(iy2,jy2);
1811 dz22 = _mm_sub_ps(iz2,jz2);
1813 /* Calculate squared distance and things based on it */
1814 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1815 rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1816 rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1817 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1818 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1819 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1820 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1821 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1822 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1824 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1825 rinv01 = gmx_mm_invsqrt_ps(rsq01);
1826 rinv02 = gmx_mm_invsqrt_ps(rsq02);
1827 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1828 rinv11 = gmx_mm_invsqrt_ps(rsq11);
1829 rinv12 = gmx_mm_invsqrt_ps(rsq12);
1830 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1831 rinv21 = gmx_mm_invsqrt_ps(rsq21);
1832 rinv22 = gmx_mm_invsqrt_ps(rsq22);
1834 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
1835 rinvsq01 = _mm_mul_ps(rinv01,rinv01);
1836 rinvsq02 = _mm_mul_ps(rinv02,rinv02);
1837 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
1838 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
1839 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
1840 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
1841 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
1842 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
1844 fjx0 = _mm_setzero_ps();
1845 fjy0 = _mm_setzero_ps();
1846 fjz0 = _mm_setzero_ps();
1847 fjx1 = _mm_setzero_ps();
1848 fjy1 = _mm_setzero_ps();
1849 fjz1 = _mm_setzero_ps();
1850 fjx2 = _mm_setzero_ps();
1851 fjy2 = _mm_setzero_ps();
1852 fjz2 = _mm_setzero_ps();
1854 /**************************
1855 * CALCULATE INTERACTIONS *
1856 **************************/
1858 if (gmx_mm_any_lt(rsq00,rcutoff2))
1861 r00 = _mm_mul_ps(rsq00,rinv00);
1863 /* EWALD ELECTROSTATICS */
1865 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1866 ewrt = _mm_mul_ps(r00,ewtabscale);
1867 ewitab = _mm_cvttps_epi32(ewrt);
1868 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1869 ewitab = _mm_slli_epi32(ewitab,2);
1870 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1871 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1872 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1873 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1874 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1875 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1876 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1877 velec = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
1878 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1880 d = _mm_sub_ps(r00,rswitch);
1881 d = _mm_max_ps(d,_mm_setzero_ps());
1882 d2 = _mm_mul_ps(d,d);
1883 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1885 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1887 /* Evaluate switch function */
1888 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1889 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1890 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
1894 fscal = _mm_and_ps(fscal,cutoff_mask);
1896 /* Calculate temporary vectorial force */
1897 tx = _mm_mul_ps(fscal,dx00);
1898 ty = _mm_mul_ps(fscal,dy00);
1899 tz = _mm_mul_ps(fscal,dz00);
1901 /* Update vectorial force */
1902 fix0 = _mm_add_ps(fix0,tx);
1903 fiy0 = _mm_add_ps(fiy0,ty);
1904 fiz0 = _mm_add_ps(fiz0,tz);
1906 fjx0 = _mm_add_ps(fjx0,tx);
1907 fjy0 = _mm_add_ps(fjy0,ty);
1908 fjz0 = _mm_add_ps(fjz0,tz);
1912 /**************************
1913 * CALCULATE INTERACTIONS *
1914 **************************/
1916 if (gmx_mm_any_lt(rsq01,rcutoff2))
1919 r01 = _mm_mul_ps(rsq01,rinv01);
1921 /* EWALD ELECTROSTATICS */
1923 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1924 ewrt = _mm_mul_ps(r01,ewtabscale);
1925 ewitab = _mm_cvttps_epi32(ewrt);
1926 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1927 ewitab = _mm_slli_epi32(ewitab,2);
1928 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1929 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1930 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1931 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1932 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1933 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1934 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1935 velec = _mm_mul_ps(qq01,_mm_sub_ps(rinv01,velec));
1936 felec = _mm_mul_ps(_mm_mul_ps(qq01,rinv01),_mm_sub_ps(rinvsq01,felec));
1938 d = _mm_sub_ps(r01,rswitch);
1939 d = _mm_max_ps(d,_mm_setzero_ps());
1940 d2 = _mm_mul_ps(d,d);
1941 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1943 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1945 /* Evaluate switch function */
1946 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1947 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv01,_mm_mul_ps(velec,dsw)) );
1948 cutoff_mask = _mm_cmplt_ps(rsq01,rcutoff2);
1952 fscal = _mm_and_ps(fscal,cutoff_mask);
1954 /* Calculate temporary vectorial force */
1955 tx = _mm_mul_ps(fscal,dx01);
1956 ty = _mm_mul_ps(fscal,dy01);
1957 tz = _mm_mul_ps(fscal,dz01);
1959 /* Update vectorial force */
1960 fix0 = _mm_add_ps(fix0,tx);
1961 fiy0 = _mm_add_ps(fiy0,ty);
1962 fiz0 = _mm_add_ps(fiz0,tz);
1964 fjx1 = _mm_add_ps(fjx1,tx);
1965 fjy1 = _mm_add_ps(fjy1,ty);
1966 fjz1 = _mm_add_ps(fjz1,tz);
1970 /**************************
1971 * CALCULATE INTERACTIONS *
1972 **************************/
1974 if (gmx_mm_any_lt(rsq02,rcutoff2))
1977 r02 = _mm_mul_ps(rsq02,rinv02);
1979 /* EWALD ELECTROSTATICS */
1981 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1982 ewrt = _mm_mul_ps(r02,ewtabscale);
1983 ewitab = _mm_cvttps_epi32(ewrt);
1984 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1985 ewitab = _mm_slli_epi32(ewitab,2);
1986 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1987 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1988 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1989 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1990 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1991 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1992 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1993 velec = _mm_mul_ps(qq02,_mm_sub_ps(rinv02,velec));
1994 felec = _mm_mul_ps(_mm_mul_ps(qq02,rinv02),_mm_sub_ps(rinvsq02,felec));
1996 d = _mm_sub_ps(r02,rswitch);
1997 d = _mm_max_ps(d,_mm_setzero_ps());
1998 d2 = _mm_mul_ps(d,d);
1999 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
2001 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
2003 /* Evaluate switch function */
2004 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2005 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv02,_mm_mul_ps(velec,dsw)) );
2006 cutoff_mask = _mm_cmplt_ps(rsq02,rcutoff2);
2010 fscal = _mm_and_ps(fscal,cutoff_mask);
2012 /* Calculate temporary vectorial force */
2013 tx = _mm_mul_ps(fscal,dx02);
2014 ty = _mm_mul_ps(fscal,dy02);
2015 tz = _mm_mul_ps(fscal,dz02);
2017 /* Update vectorial force */
2018 fix0 = _mm_add_ps(fix0,tx);
2019 fiy0 = _mm_add_ps(fiy0,ty);
2020 fiz0 = _mm_add_ps(fiz0,tz);
2022 fjx2 = _mm_add_ps(fjx2,tx);
2023 fjy2 = _mm_add_ps(fjy2,ty);
2024 fjz2 = _mm_add_ps(fjz2,tz);
2028 /**************************
2029 * CALCULATE INTERACTIONS *
2030 **************************/
2032 if (gmx_mm_any_lt(rsq10,rcutoff2))
2035 r10 = _mm_mul_ps(rsq10,rinv10);
2037 /* EWALD ELECTROSTATICS */
2039 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2040 ewrt = _mm_mul_ps(r10,ewtabscale);
2041 ewitab = _mm_cvttps_epi32(ewrt);
2042 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2043 ewitab = _mm_slli_epi32(ewitab,2);
2044 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2045 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2046 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
2047 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
2048 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
2049 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
2050 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
2051 velec = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
2052 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
2054 d = _mm_sub_ps(r10,rswitch);
2055 d = _mm_max_ps(d,_mm_setzero_ps());
2056 d2 = _mm_mul_ps(d,d);
2057 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
2059 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
2061 /* Evaluate switch function */
2062 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2063 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
2064 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
2068 fscal = _mm_and_ps(fscal,cutoff_mask);
2070 /* Calculate temporary vectorial force */
2071 tx = _mm_mul_ps(fscal,dx10);
2072 ty = _mm_mul_ps(fscal,dy10);
2073 tz = _mm_mul_ps(fscal,dz10);
2075 /* Update vectorial force */
2076 fix1 = _mm_add_ps(fix1,tx);
2077 fiy1 = _mm_add_ps(fiy1,ty);
2078 fiz1 = _mm_add_ps(fiz1,tz);
2080 fjx0 = _mm_add_ps(fjx0,tx);
2081 fjy0 = _mm_add_ps(fjy0,ty);
2082 fjz0 = _mm_add_ps(fjz0,tz);
2086 /**************************
2087 * CALCULATE INTERACTIONS *
2088 **************************/
2090 if (gmx_mm_any_lt(rsq11,rcutoff2))
2093 r11 = _mm_mul_ps(rsq11,rinv11);
2095 /* EWALD ELECTROSTATICS */
2097 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2098 ewrt = _mm_mul_ps(r11,ewtabscale);
2099 ewitab = _mm_cvttps_epi32(ewrt);
2100 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2101 ewitab = _mm_slli_epi32(ewitab,2);
2102 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2103 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2104 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
2105 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
2106 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
2107 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
2108 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
2109 velec = _mm_mul_ps(qq11,_mm_sub_ps(rinv11,velec));
2110 felec = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
2112 d = _mm_sub_ps(r11,rswitch);
2113 d = _mm_max_ps(d,_mm_setzero_ps());
2114 d2 = _mm_mul_ps(d,d);
2115 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
2117 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
2119 /* Evaluate switch function */
2120 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2121 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv11,_mm_mul_ps(velec,dsw)) );
2122 cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2);
2126 fscal = _mm_and_ps(fscal,cutoff_mask);
2128 /* Calculate temporary vectorial force */
2129 tx = _mm_mul_ps(fscal,dx11);
2130 ty = _mm_mul_ps(fscal,dy11);
2131 tz = _mm_mul_ps(fscal,dz11);
2133 /* Update vectorial force */
2134 fix1 = _mm_add_ps(fix1,tx);
2135 fiy1 = _mm_add_ps(fiy1,ty);
2136 fiz1 = _mm_add_ps(fiz1,tz);
2138 fjx1 = _mm_add_ps(fjx1,tx);
2139 fjy1 = _mm_add_ps(fjy1,ty);
2140 fjz1 = _mm_add_ps(fjz1,tz);
2144 /**************************
2145 * CALCULATE INTERACTIONS *
2146 **************************/
2148 if (gmx_mm_any_lt(rsq12,rcutoff2))
2151 r12 = _mm_mul_ps(rsq12,rinv12);
2153 /* EWALD ELECTROSTATICS */
2155 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2156 ewrt = _mm_mul_ps(r12,ewtabscale);
2157 ewitab = _mm_cvttps_epi32(ewrt);
2158 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2159 ewitab = _mm_slli_epi32(ewitab,2);
2160 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2161 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2162 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
2163 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
2164 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
2165 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
2166 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
2167 velec = _mm_mul_ps(qq12,_mm_sub_ps(rinv12,velec));
2168 felec = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
2170 d = _mm_sub_ps(r12,rswitch);
2171 d = _mm_max_ps(d,_mm_setzero_ps());
2172 d2 = _mm_mul_ps(d,d);
2173 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
2175 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
2177 /* Evaluate switch function */
2178 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2179 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv12,_mm_mul_ps(velec,dsw)) );
2180 cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2);
2184 fscal = _mm_and_ps(fscal,cutoff_mask);
2186 /* Calculate temporary vectorial force */
2187 tx = _mm_mul_ps(fscal,dx12);
2188 ty = _mm_mul_ps(fscal,dy12);
2189 tz = _mm_mul_ps(fscal,dz12);
2191 /* Update vectorial force */
2192 fix1 = _mm_add_ps(fix1,tx);
2193 fiy1 = _mm_add_ps(fiy1,ty);
2194 fiz1 = _mm_add_ps(fiz1,tz);
2196 fjx2 = _mm_add_ps(fjx2,tx);
2197 fjy2 = _mm_add_ps(fjy2,ty);
2198 fjz2 = _mm_add_ps(fjz2,tz);
2202 /**************************
2203 * CALCULATE INTERACTIONS *
2204 **************************/
2206 if (gmx_mm_any_lt(rsq20,rcutoff2))
2209 r20 = _mm_mul_ps(rsq20,rinv20);
2211 /* EWALD ELECTROSTATICS */
2213 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2214 ewrt = _mm_mul_ps(r20,ewtabscale);
2215 ewitab = _mm_cvttps_epi32(ewrt);
2216 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2217 ewitab = _mm_slli_epi32(ewitab,2);
2218 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2219 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2220 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
2221 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
2222 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
2223 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
2224 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
2225 velec = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
2226 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
2228 d = _mm_sub_ps(r20,rswitch);
2229 d = _mm_max_ps(d,_mm_setzero_ps());
2230 d2 = _mm_mul_ps(d,d);
2231 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
2233 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
2235 /* Evaluate switch function */
2236 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2237 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
2238 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
2242 fscal = _mm_and_ps(fscal,cutoff_mask);
2244 /* Calculate temporary vectorial force */
2245 tx = _mm_mul_ps(fscal,dx20);
2246 ty = _mm_mul_ps(fscal,dy20);
2247 tz = _mm_mul_ps(fscal,dz20);
2249 /* Update vectorial force */
2250 fix2 = _mm_add_ps(fix2,tx);
2251 fiy2 = _mm_add_ps(fiy2,ty);
2252 fiz2 = _mm_add_ps(fiz2,tz);
2254 fjx0 = _mm_add_ps(fjx0,tx);
2255 fjy0 = _mm_add_ps(fjy0,ty);
2256 fjz0 = _mm_add_ps(fjz0,tz);
2260 /**************************
2261 * CALCULATE INTERACTIONS *
2262 **************************/
2264 if (gmx_mm_any_lt(rsq21,rcutoff2))
2267 r21 = _mm_mul_ps(rsq21,rinv21);
2269 /* EWALD ELECTROSTATICS */
2271 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2272 ewrt = _mm_mul_ps(r21,ewtabscale);
2273 ewitab = _mm_cvttps_epi32(ewrt);
2274 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2275 ewitab = _mm_slli_epi32(ewitab,2);
2276 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2277 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2278 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
2279 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
2280 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
2281 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
2282 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
2283 velec = _mm_mul_ps(qq21,_mm_sub_ps(rinv21,velec));
2284 felec = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
2286 d = _mm_sub_ps(r21,rswitch);
2287 d = _mm_max_ps(d,_mm_setzero_ps());
2288 d2 = _mm_mul_ps(d,d);
2289 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
2291 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
2293 /* Evaluate switch function */
2294 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2295 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv21,_mm_mul_ps(velec,dsw)) );
2296 cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2);
2300 fscal = _mm_and_ps(fscal,cutoff_mask);
2302 /* Calculate temporary vectorial force */
2303 tx = _mm_mul_ps(fscal,dx21);
2304 ty = _mm_mul_ps(fscal,dy21);
2305 tz = _mm_mul_ps(fscal,dz21);
2307 /* Update vectorial force */
2308 fix2 = _mm_add_ps(fix2,tx);
2309 fiy2 = _mm_add_ps(fiy2,ty);
2310 fiz2 = _mm_add_ps(fiz2,tz);
2312 fjx1 = _mm_add_ps(fjx1,tx);
2313 fjy1 = _mm_add_ps(fjy1,ty);
2314 fjz1 = _mm_add_ps(fjz1,tz);
2318 /**************************
2319 * CALCULATE INTERACTIONS *
2320 **************************/
2322 if (gmx_mm_any_lt(rsq22,rcutoff2))
2325 r22 = _mm_mul_ps(rsq22,rinv22);
2327 /* EWALD ELECTROSTATICS */
2329 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2330 ewrt = _mm_mul_ps(r22,ewtabscale);
2331 ewitab = _mm_cvttps_epi32(ewrt);
2332 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2333 ewitab = _mm_slli_epi32(ewitab,2);
2334 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2335 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2336 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
2337 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
2338 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
2339 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
2340 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
2341 velec = _mm_mul_ps(qq22,_mm_sub_ps(rinv22,velec));
2342 felec = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
2344 d = _mm_sub_ps(r22,rswitch);
2345 d = _mm_max_ps(d,_mm_setzero_ps());
2346 d2 = _mm_mul_ps(d,d);
2347 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
2349 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
2351 /* Evaluate switch function */
2352 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2353 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv22,_mm_mul_ps(velec,dsw)) );
2354 cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2);
2358 fscal = _mm_and_ps(fscal,cutoff_mask);
2360 /* Calculate temporary vectorial force */
2361 tx = _mm_mul_ps(fscal,dx22);
2362 ty = _mm_mul_ps(fscal,dy22);
2363 tz = _mm_mul_ps(fscal,dz22);
2365 /* Update vectorial force */
2366 fix2 = _mm_add_ps(fix2,tx);
2367 fiy2 = _mm_add_ps(fiy2,ty);
2368 fiz2 = _mm_add_ps(fiz2,tz);
2370 fjx2 = _mm_add_ps(fjx2,tx);
2371 fjy2 = _mm_add_ps(fjy2,ty);
2372 fjz2 = _mm_add_ps(fjz2,tz);
2376 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
2377 f+j_coord_offsetC,f+j_coord_offsetD,
2378 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2380 /* Inner loop uses 558 flops */
2383 if(jidx<j_index_end)
2386 /* Get j neighbor index, and coordinate index */
2388 jnrB = jjnr[jidx+1];
2389 jnrC = jjnr[jidx+2];
2390 jnrD = jjnr[jidx+3];
2392 /* Sign of each element will be negative for non-real atoms.
2393 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2394 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2396 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2397 jnrA = (jnrA>=0) ? jnrA : 0;
2398 jnrB = (jnrB>=0) ? jnrB : 0;
2399 jnrC = (jnrC>=0) ? jnrC : 0;
2400 jnrD = (jnrD>=0) ? jnrD : 0;
2402 j_coord_offsetA = DIM*jnrA;
2403 j_coord_offsetB = DIM*jnrB;
2404 j_coord_offsetC = DIM*jnrC;
2405 j_coord_offsetD = DIM*jnrD;
2407 /* load j atom coordinates */
2408 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2409 x+j_coord_offsetC,x+j_coord_offsetD,
2410 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
2412 /* Calculate displacement vector */
2413 dx00 = _mm_sub_ps(ix0,jx0);
2414 dy00 = _mm_sub_ps(iy0,jy0);
2415 dz00 = _mm_sub_ps(iz0,jz0);
2416 dx01 = _mm_sub_ps(ix0,jx1);
2417 dy01 = _mm_sub_ps(iy0,jy1);
2418 dz01 = _mm_sub_ps(iz0,jz1);
2419 dx02 = _mm_sub_ps(ix0,jx2);
2420 dy02 = _mm_sub_ps(iy0,jy2);
2421 dz02 = _mm_sub_ps(iz0,jz2);
2422 dx10 = _mm_sub_ps(ix1,jx0);
2423 dy10 = _mm_sub_ps(iy1,jy0);
2424 dz10 = _mm_sub_ps(iz1,jz0);
2425 dx11 = _mm_sub_ps(ix1,jx1);
2426 dy11 = _mm_sub_ps(iy1,jy1);
2427 dz11 = _mm_sub_ps(iz1,jz1);
2428 dx12 = _mm_sub_ps(ix1,jx2);
2429 dy12 = _mm_sub_ps(iy1,jy2);
2430 dz12 = _mm_sub_ps(iz1,jz2);
2431 dx20 = _mm_sub_ps(ix2,jx0);
2432 dy20 = _mm_sub_ps(iy2,jy0);
2433 dz20 = _mm_sub_ps(iz2,jz0);
2434 dx21 = _mm_sub_ps(ix2,jx1);
2435 dy21 = _mm_sub_ps(iy2,jy1);
2436 dz21 = _mm_sub_ps(iz2,jz1);
2437 dx22 = _mm_sub_ps(ix2,jx2);
2438 dy22 = _mm_sub_ps(iy2,jy2);
2439 dz22 = _mm_sub_ps(iz2,jz2);
2441 /* Calculate squared distance and things based on it */
2442 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
2443 rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
2444 rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
2445 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
2446 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
2447 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
2448 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
2449 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
2450 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
2452 rinv00 = gmx_mm_invsqrt_ps(rsq00);
2453 rinv01 = gmx_mm_invsqrt_ps(rsq01);
2454 rinv02 = gmx_mm_invsqrt_ps(rsq02);
2455 rinv10 = gmx_mm_invsqrt_ps(rsq10);
2456 rinv11 = gmx_mm_invsqrt_ps(rsq11);
2457 rinv12 = gmx_mm_invsqrt_ps(rsq12);
2458 rinv20 = gmx_mm_invsqrt_ps(rsq20);
2459 rinv21 = gmx_mm_invsqrt_ps(rsq21);
2460 rinv22 = gmx_mm_invsqrt_ps(rsq22);
2462 rinvsq00 = _mm_mul_ps(rinv00,rinv00);
2463 rinvsq01 = _mm_mul_ps(rinv01,rinv01);
2464 rinvsq02 = _mm_mul_ps(rinv02,rinv02);
2465 rinvsq10 = _mm_mul_ps(rinv10,rinv10);
2466 rinvsq11 = _mm_mul_ps(rinv11,rinv11);
2467 rinvsq12 = _mm_mul_ps(rinv12,rinv12);
2468 rinvsq20 = _mm_mul_ps(rinv20,rinv20);
2469 rinvsq21 = _mm_mul_ps(rinv21,rinv21);
2470 rinvsq22 = _mm_mul_ps(rinv22,rinv22);
2472 fjx0 = _mm_setzero_ps();
2473 fjy0 = _mm_setzero_ps();
2474 fjz0 = _mm_setzero_ps();
2475 fjx1 = _mm_setzero_ps();
2476 fjy1 = _mm_setzero_ps();
2477 fjz1 = _mm_setzero_ps();
2478 fjx2 = _mm_setzero_ps();
2479 fjy2 = _mm_setzero_ps();
2480 fjz2 = _mm_setzero_ps();
2482 /**************************
2483 * CALCULATE INTERACTIONS *
2484 **************************/
2486 if (gmx_mm_any_lt(rsq00,rcutoff2))
2489 r00 = _mm_mul_ps(rsq00,rinv00);
2490 r00 = _mm_andnot_ps(dummy_mask,r00);
2492 /* EWALD ELECTROSTATICS */
2494 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2495 ewrt = _mm_mul_ps(r00,ewtabscale);
2496 ewitab = _mm_cvttps_epi32(ewrt);
2497 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2498 ewitab = _mm_slli_epi32(ewitab,2);
2499 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2500 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2501 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
2502 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
2503 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
2504 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
2505 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
2506 velec = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
2507 felec = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
2509 d = _mm_sub_ps(r00,rswitch);
2510 d = _mm_max_ps(d,_mm_setzero_ps());
2511 d2 = _mm_mul_ps(d,d);
2512 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
2514 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
2516 /* Evaluate switch function */
2517 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2518 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
2519 cutoff_mask = _mm_cmplt_ps(rsq00,rcutoff2);
2523 fscal = _mm_and_ps(fscal,cutoff_mask);
2525 fscal = _mm_andnot_ps(dummy_mask,fscal);
2527 /* Calculate temporary vectorial force */
2528 tx = _mm_mul_ps(fscal,dx00);
2529 ty = _mm_mul_ps(fscal,dy00);
2530 tz = _mm_mul_ps(fscal,dz00);
2532 /* Update vectorial force */
2533 fix0 = _mm_add_ps(fix0,tx);
2534 fiy0 = _mm_add_ps(fiy0,ty);
2535 fiz0 = _mm_add_ps(fiz0,tz);
2537 fjx0 = _mm_add_ps(fjx0,tx);
2538 fjy0 = _mm_add_ps(fjy0,ty);
2539 fjz0 = _mm_add_ps(fjz0,tz);
2543 /**************************
2544 * CALCULATE INTERACTIONS *
2545 **************************/
2547 if (gmx_mm_any_lt(rsq01,rcutoff2))
2550 r01 = _mm_mul_ps(rsq01,rinv01);
2551 r01 = _mm_andnot_ps(dummy_mask,r01);
2553 /* EWALD ELECTROSTATICS */
2555 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2556 ewrt = _mm_mul_ps(r01,ewtabscale);
2557 ewitab = _mm_cvttps_epi32(ewrt);
2558 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2559 ewitab = _mm_slli_epi32(ewitab,2);
2560 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2561 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2562 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
2563 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
2564 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
2565 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
2566 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
2567 velec = _mm_mul_ps(qq01,_mm_sub_ps(rinv01,velec));
2568 felec = _mm_mul_ps(_mm_mul_ps(qq01,rinv01),_mm_sub_ps(rinvsq01,felec));
2570 d = _mm_sub_ps(r01,rswitch);
2571 d = _mm_max_ps(d,_mm_setzero_ps());
2572 d2 = _mm_mul_ps(d,d);
2573 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
2575 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
2577 /* Evaluate switch function */
2578 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2579 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv01,_mm_mul_ps(velec,dsw)) );
2580 cutoff_mask = _mm_cmplt_ps(rsq01,rcutoff2);
2584 fscal = _mm_and_ps(fscal,cutoff_mask);
2586 fscal = _mm_andnot_ps(dummy_mask,fscal);
2588 /* Calculate temporary vectorial force */
2589 tx = _mm_mul_ps(fscal,dx01);
2590 ty = _mm_mul_ps(fscal,dy01);
2591 tz = _mm_mul_ps(fscal,dz01);
2593 /* Update vectorial force */
2594 fix0 = _mm_add_ps(fix0,tx);
2595 fiy0 = _mm_add_ps(fiy0,ty);
2596 fiz0 = _mm_add_ps(fiz0,tz);
2598 fjx1 = _mm_add_ps(fjx1,tx);
2599 fjy1 = _mm_add_ps(fjy1,ty);
2600 fjz1 = _mm_add_ps(fjz1,tz);
2604 /**************************
2605 * CALCULATE INTERACTIONS *
2606 **************************/
2608 if (gmx_mm_any_lt(rsq02,rcutoff2))
2611 r02 = _mm_mul_ps(rsq02,rinv02);
2612 r02 = _mm_andnot_ps(dummy_mask,r02);
2614 /* EWALD ELECTROSTATICS */
2616 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2617 ewrt = _mm_mul_ps(r02,ewtabscale);
2618 ewitab = _mm_cvttps_epi32(ewrt);
2619 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2620 ewitab = _mm_slli_epi32(ewitab,2);
2621 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2622 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2623 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
2624 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
2625 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
2626 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
2627 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
2628 velec = _mm_mul_ps(qq02,_mm_sub_ps(rinv02,velec));
2629 felec = _mm_mul_ps(_mm_mul_ps(qq02,rinv02),_mm_sub_ps(rinvsq02,felec));
2631 d = _mm_sub_ps(r02,rswitch);
2632 d = _mm_max_ps(d,_mm_setzero_ps());
2633 d2 = _mm_mul_ps(d,d);
2634 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
2636 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
2638 /* Evaluate switch function */
2639 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2640 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv02,_mm_mul_ps(velec,dsw)) );
2641 cutoff_mask = _mm_cmplt_ps(rsq02,rcutoff2);
2645 fscal = _mm_and_ps(fscal,cutoff_mask);
2647 fscal = _mm_andnot_ps(dummy_mask,fscal);
2649 /* Calculate temporary vectorial force */
2650 tx = _mm_mul_ps(fscal,dx02);
2651 ty = _mm_mul_ps(fscal,dy02);
2652 tz = _mm_mul_ps(fscal,dz02);
2654 /* Update vectorial force */
2655 fix0 = _mm_add_ps(fix0,tx);
2656 fiy0 = _mm_add_ps(fiy0,ty);
2657 fiz0 = _mm_add_ps(fiz0,tz);
2659 fjx2 = _mm_add_ps(fjx2,tx);
2660 fjy2 = _mm_add_ps(fjy2,ty);
2661 fjz2 = _mm_add_ps(fjz2,tz);
2665 /**************************
2666 * CALCULATE INTERACTIONS *
2667 **************************/
2669 if (gmx_mm_any_lt(rsq10,rcutoff2))
2672 r10 = _mm_mul_ps(rsq10,rinv10);
2673 r10 = _mm_andnot_ps(dummy_mask,r10);
2675 /* EWALD ELECTROSTATICS */
2677 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2678 ewrt = _mm_mul_ps(r10,ewtabscale);
2679 ewitab = _mm_cvttps_epi32(ewrt);
2680 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2681 ewitab = _mm_slli_epi32(ewitab,2);
2682 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2683 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2684 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
2685 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
2686 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
2687 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
2688 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
2689 velec = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
2690 felec = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
2692 d = _mm_sub_ps(r10,rswitch);
2693 d = _mm_max_ps(d,_mm_setzero_ps());
2694 d2 = _mm_mul_ps(d,d);
2695 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
2697 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
2699 /* Evaluate switch function */
2700 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2701 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
2702 cutoff_mask = _mm_cmplt_ps(rsq10,rcutoff2);
2706 fscal = _mm_and_ps(fscal,cutoff_mask);
2708 fscal = _mm_andnot_ps(dummy_mask,fscal);
2710 /* Calculate temporary vectorial force */
2711 tx = _mm_mul_ps(fscal,dx10);
2712 ty = _mm_mul_ps(fscal,dy10);
2713 tz = _mm_mul_ps(fscal,dz10);
2715 /* Update vectorial force */
2716 fix1 = _mm_add_ps(fix1,tx);
2717 fiy1 = _mm_add_ps(fiy1,ty);
2718 fiz1 = _mm_add_ps(fiz1,tz);
2720 fjx0 = _mm_add_ps(fjx0,tx);
2721 fjy0 = _mm_add_ps(fjy0,ty);
2722 fjz0 = _mm_add_ps(fjz0,tz);
2726 /**************************
2727 * CALCULATE INTERACTIONS *
2728 **************************/
2730 if (gmx_mm_any_lt(rsq11,rcutoff2))
2733 r11 = _mm_mul_ps(rsq11,rinv11);
2734 r11 = _mm_andnot_ps(dummy_mask,r11);
2736 /* EWALD ELECTROSTATICS */
2738 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2739 ewrt = _mm_mul_ps(r11,ewtabscale);
2740 ewitab = _mm_cvttps_epi32(ewrt);
2741 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2742 ewitab = _mm_slli_epi32(ewitab,2);
2743 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2744 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2745 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
2746 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
2747 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
2748 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
2749 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
2750 velec = _mm_mul_ps(qq11,_mm_sub_ps(rinv11,velec));
2751 felec = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
2753 d = _mm_sub_ps(r11,rswitch);
2754 d = _mm_max_ps(d,_mm_setzero_ps());
2755 d2 = _mm_mul_ps(d,d);
2756 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
2758 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
2760 /* Evaluate switch function */
2761 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2762 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv11,_mm_mul_ps(velec,dsw)) );
2763 cutoff_mask = _mm_cmplt_ps(rsq11,rcutoff2);
2767 fscal = _mm_and_ps(fscal,cutoff_mask);
2769 fscal = _mm_andnot_ps(dummy_mask,fscal);
2771 /* Calculate temporary vectorial force */
2772 tx = _mm_mul_ps(fscal,dx11);
2773 ty = _mm_mul_ps(fscal,dy11);
2774 tz = _mm_mul_ps(fscal,dz11);
2776 /* Update vectorial force */
2777 fix1 = _mm_add_ps(fix1,tx);
2778 fiy1 = _mm_add_ps(fiy1,ty);
2779 fiz1 = _mm_add_ps(fiz1,tz);
2781 fjx1 = _mm_add_ps(fjx1,tx);
2782 fjy1 = _mm_add_ps(fjy1,ty);
2783 fjz1 = _mm_add_ps(fjz1,tz);
2787 /**************************
2788 * CALCULATE INTERACTIONS *
2789 **************************/
2791 if (gmx_mm_any_lt(rsq12,rcutoff2))
2794 r12 = _mm_mul_ps(rsq12,rinv12);
2795 r12 = _mm_andnot_ps(dummy_mask,r12);
2797 /* EWALD ELECTROSTATICS */
2799 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2800 ewrt = _mm_mul_ps(r12,ewtabscale);
2801 ewitab = _mm_cvttps_epi32(ewrt);
2802 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2803 ewitab = _mm_slli_epi32(ewitab,2);
2804 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2805 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2806 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
2807 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
2808 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
2809 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
2810 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
2811 velec = _mm_mul_ps(qq12,_mm_sub_ps(rinv12,velec));
2812 felec = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
2814 d = _mm_sub_ps(r12,rswitch);
2815 d = _mm_max_ps(d,_mm_setzero_ps());
2816 d2 = _mm_mul_ps(d,d);
2817 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
2819 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
2821 /* Evaluate switch function */
2822 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2823 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv12,_mm_mul_ps(velec,dsw)) );
2824 cutoff_mask = _mm_cmplt_ps(rsq12,rcutoff2);
2828 fscal = _mm_and_ps(fscal,cutoff_mask);
2830 fscal = _mm_andnot_ps(dummy_mask,fscal);
2832 /* Calculate temporary vectorial force */
2833 tx = _mm_mul_ps(fscal,dx12);
2834 ty = _mm_mul_ps(fscal,dy12);
2835 tz = _mm_mul_ps(fscal,dz12);
2837 /* Update vectorial force */
2838 fix1 = _mm_add_ps(fix1,tx);
2839 fiy1 = _mm_add_ps(fiy1,ty);
2840 fiz1 = _mm_add_ps(fiz1,tz);
2842 fjx2 = _mm_add_ps(fjx2,tx);
2843 fjy2 = _mm_add_ps(fjy2,ty);
2844 fjz2 = _mm_add_ps(fjz2,tz);
2848 /**************************
2849 * CALCULATE INTERACTIONS *
2850 **************************/
2852 if (gmx_mm_any_lt(rsq20,rcutoff2))
2855 r20 = _mm_mul_ps(rsq20,rinv20);
2856 r20 = _mm_andnot_ps(dummy_mask,r20);
2858 /* EWALD ELECTROSTATICS */
2860 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2861 ewrt = _mm_mul_ps(r20,ewtabscale);
2862 ewitab = _mm_cvttps_epi32(ewrt);
2863 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2864 ewitab = _mm_slli_epi32(ewitab,2);
2865 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2866 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2867 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
2868 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
2869 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
2870 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
2871 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
2872 velec = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
2873 felec = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
2875 d = _mm_sub_ps(r20,rswitch);
2876 d = _mm_max_ps(d,_mm_setzero_ps());
2877 d2 = _mm_mul_ps(d,d);
2878 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
2880 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
2882 /* Evaluate switch function */
2883 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2884 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
2885 cutoff_mask = _mm_cmplt_ps(rsq20,rcutoff2);
2889 fscal = _mm_and_ps(fscal,cutoff_mask);
2891 fscal = _mm_andnot_ps(dummy_mask,fscal);
2893 /* Calculate temporary vectorial force */
2894 tx = _mm_mul_ps(fscal,dx20);
2895 ty = _mm_mul_ps(fscal,dy20);
2896 tz = _mm_mul_ps(fscal,dz20);
2898 /* Update vectorial force */
2899 fix2 = _mm_add_ps(fix2,tx);
2900 fiy2 = _mm_add_ps(fiy2,ty);
2901 fiz2 = _mm_add_ps(fiz2,tz);
2903 fjx0 = _mm_add_ps(fjx0,tx);
2904 fjy0 = _mm_add_ps(fjy0,ty);
2905 fjz0 = _mm_add_ps(fjz0,tz);
2909 /**************************
2910 * CALCULATE INTERACTIONS *
2911 **************************/
2913 if (gmx_mm_any_lt(rsq21,rcutoff2))
2916 r21 = _mm_mul_ps(rsq21,rinv21);
2917 r21 = _mm_andnot_ps(dummy_mask,r21);
2919 /* EWALD ELECTROSTATICS */
2921 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2922 ewrt = _mm_mul_ps(r21,ewtabscale);
2923 ewitab = _mm_cvttps_epi32(ewrt);
2924 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2925 ewitab = _mm_slli_epi32(ewitab,2);
2926 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2927 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2928 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
2929 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
2930 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
2931 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
2932 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
2933 velec = _mm_mul_ps(qq21,_mm_sub_ps(rinv21,velec));
2934 felec = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
2936 d = _mm_sub_ps(r21,rswitch);
2937 d = _mm_max_ps(d,_mm_setzero_ps());
2938 d2 = _mm_mul_ps(d,d);
2939 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
2941 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
2943 /* Evaluate switch function */
2944 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
2945 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv21,_mm_mul_ps(velec,dsw)) );
2946 cutoff_mask = _mm_cmplt_ps(rsq21,rcutoff2);
2950 fscal = _mm_and_ps(fscal,cutoff_mask);
2952 fscal = _mm_andnot_ps(dummy_mask,fscal);
2954 /* Calculate temporary vectorial force */
2955 tx = _mm_mul_ps(fscal,dx21);
2956 ty = _mm_mul_ps(fscal,dy21);
2957 tz = _mm_mul_ps(fscal,dz21);
2959 /* Update vectorial force */
2960 fix2 = _mm_add_ps(fix2,tx);
2961 fiy2 = _mm_add_ps(fiy2,ty);
2962 fiz2 = _mm_add_ps(fiz2,tz);
2964 fjx1 = _mm_add_ps(fjx1,tx);
2965 fjy1 = _mm_add_ps(fjy1,ty);
2966 fjz1 = _mm_add_ps(fjz1,tz);
2970 /**************************
2971 * CALCULATE INTERACTIONS *
2972 **************************/
2974 if (gmx_mm_any_lt(rsq22,rcutoff2))
2977 r22 = _mm_mul_ps(rsq22,rinv22);
2978 r22 = _mm_andnot_ps(dummy_mask,r22);
2980 /* EWALD ELECTROSTATICS */
2982 /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2983 ewrt = _mm_mul_ps(r22,ewtabscale);
2984 ewitab = _mm_cvttps_epi32(ewrt);
2985 eweps = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
2986 ewitab = _mm_slli_epi32(ewitab,2);
2987 ewtabF = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
2988 ewtabD = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
2989 ewtabV = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
2990 ewtabFn = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
2991 _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
2992 felec = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
2993 velec = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
2994 velec = _mm_mul_ps(qq22,_mm_sub_ps(rinv22,velec));
2995 felec = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
2997 d = _mm_sub_ps(r22,rswitch);
2998 d = _mm_max_ps(d,_mm_setzero_ps());
2999 d2 = _mm_mul_ps(d,d);
3000 sw = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
3002 dsw = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
3004 /* Evaluate switch function */
3005 /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
3006 felec = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv22,_mm_mul_ps(velec,dsw)) );
3007 cutoff_mask = _mm_cmplt_ps(rsq22,rcutoff2);
3011 fscal = _mm_and_ps(fscal,cutoff_mask);
3013 fscal = _mm_andnot_ps(dummy_mask,fscal);
3015 /* Calculate temporary vectorial force */
3016 tx = _mm_mul_ps(fscal,dx22);
3017 ty = _mm_mul_ps(fscal,dy22);
3018 tz = _mm_mul_ps(fscal,dz22);
3020 /* Update vectorial force */
3021 fix2 = _mm_add_ps(fix2,tx);
3022 fiy2 = _mm_add_ps(fiy2,ty);
3023 fiz2 = _mm_add_ps(fiz2,tz);
3025 fjx2 = _mm_add_ps(fjx2,tx);
3026 fjy2 = _mm_add_ps(fjy2,ty);
3027 fjz2 = _mm_add_ps(fjz2,tz);
3031 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
3032 f+j_coord_offsetC,f+j_coord_offsetD,
3033 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
3035 /* Inner loop uses 567 flops */
3038 /* End of innermost loop */
3040 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
3041 f+i_coord_offset,fshift+i_shift_offset);
3043 /* Increment number of inner iterations */
3044 inneriter += j_index_end - j_index_start;
3046 /* Outer loop uses 27 flops */
3049 /* Increment number of outer iterations */
3052 /* Update outer/inner flops */
3054 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*27 + inneriter*567);