2 * Note: this file was generated by the Gromacs sse2_single kernel generator.
4 * This source code is part of
8 * Copyright (c) 2001-2012, The GROMACS Development Team
10 * Gromacs is a library for molecular simulation and trajectory analysis,
11 * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
12 * a full list of developers and information, check out http://www.gromacs.org
14 * This program is free software; you can redistribute it and/or modify it under
15 * the terms of the GNU Lesser General Public License as published by the Free
16 * Software Foundation; either version 2 of the License, or (at your option) any
19 * To help fund GROMACS development, we humbly ask that you cite
20 * the papers people have written on it - you can find them on the website.
28 #include "../nb_kernel.h"
29 #include "types/simple.h"
33 #include "gmx_math_x86_sse2_single.h"
34 #include "kernelutil_x86_sse2_single.h"
37 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW3W3_VF_sse2_single
38 * Electrostatics interaction: CubicSplineTable
39 * VdW interaction: None
40 * Geometry: Water3-Water3
41 * Calculate force/pot: PotentialAndForce
44 nb_kernel_ElecCSTab_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 __m128i ifour = _mm_set1_epi32(4);
91 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
93 __m128 dummy_mask,cutoff_mask;
94 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
95 __m128 one = _mm_set1_ps(1.0);
96 __m128 two = _mm_set1_ps(2.0);
102 jindex = nlist->jindex;
104 shiftidx = nlist->shift;
106 shiftvec = fr->shift_vec[0];
107 fshift = fr->fshift[0];
108 facel = _mm_set1_ps(fr->epsfac);
109 charge = mdatoms->chargeA;
111 vftab = kernel_data->table_elec->data;
112 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
114 /* Setup water-specific parameters */
115 inr = nlist->iinr[0];
116 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
117 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
118 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
120 jq0 = _mm_set1_ps(charge[inr+0]);
121 jq1 = _mm_set1_ps(charge[inr+1]);
122 jq2 = _mm_set1_ps(charge[inr+2]);
123 qq00 = _mm_mul_ps(iq0,jq0);
124 qq01 = _mm_mul_ps(iq0,jq1);
125 qq02 = _mm_mul_ps(iq0,jq2);
126 qq10 = _mm_mul_ps(iq1,jq0);
127 qq11 = _mm_mul_ps(iq1,jq1);
128 qq12 = _mm_mul_ps(iq1,jq2);
129 qq20 = _mm_mul_ps(iq2,jq0);
130 qq21 = _mm_mul_ps(iq2,jq1);
131 qq22 = _mm_mul_ps(iq2,jq2);
133 /* Avoid stupid compiler warnings */
134 jnrA = jnrB = jnrC = jnrD = 0;
143 /* Start outer loop over neighborlists */
144 for(iidx=0; iidx<nri; iidx++)
146 /* Load shift vector for this list */
147 i_shift_offset = DIM*shiftidx[iidx];
148 shX = shiftvec[i_shift_offset+XX];
149 shY = shiftvec[i_shift_offset+YY];
150 shZ = shiftvec[i_shift_offset+ZZ];
152 /* Load limits for loop over neighbors */
153 j_index_start = jindex[iidx];
154 j_index_end = jindex[iidx+1];
156 /* Get outer coordinate index */
158 i_coord_offset = DIM*inr;
160 /* Load i particle coords and add shift vector */
161 ix0 = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
162 iy0 = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
163 iz0 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
164 ix1 = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
165 iy1 = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
166 iz1 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
167 ix2 = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
168 iy2 = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
169 iz2 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
171 fix0 = _mm_setzero_ps();
172 fiy0 = _mm_setzero_ps();
173 fiz0 = _mm_setzero_ps();
174 fix1 = _mm_setzero_ps();
175 fiy1 = _mm_setzero_ps();
176 fiz1 = _mm_setzero_ps();
177 fix2 = _mm_setzero_ps();
178 fiy2 = _mm_setzero_ps();
179 fiz2 = _mm_setzero_ps();
181 /* Reset potential sums */
182 velecsum = _mm_setzero_ps();
184 /* Start inner kernel loop */
185 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
188 /* Get j neighbor index, and coordinate index */
194 j_coord_offsetA = DIM*jnrA;
195 j_coord_offsetB = DIM*jnrB;
196 j_coord_offsetC = DIM*jnrC;
197 j_coord_offsetD = DIM*jnrD;
199 /* load j atom coordinates */
200 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
201 x+j_coord_offsetC,x+j_coord_offsetD,
202 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
204 /* Calculate displacement vector */
205 dx00 = _mm_sub_ps(ix0,jx0);
206 dy00 = _mm_sub_ps(iy0,jy0);
207 dz00 = _mm_sub_ps(iz0,jz0);
208 dx01 = _mm_sub_ps(ix0,jx1);
209 dy01 = _mm_sub_ps(iy0,jy1);
210 dz01 = _mm_sub_ps(iz0,jz1);
211 dx02 = _mm_sub_ps(ix0,jx2);
212 dy02 = _mm_sub_ps(iy0,jy2);
213 dz02 = _mm_sub_ps(iz0,jz2);
214 dx10 = _mm_sub_ps(ix1,jx0);
215 dy10 = _mm_sub_ps(iy1,jy0);
216 dz10 = _mm_sub_ps(iz1,jz0);
217 dx11 = _mm_sub_ps(ix1,jx1);
218 dy11 = _mm_sub_ps(iy1,jy1);
219 dz11 = _mm_sub_ps(iz1,jz1);
220 dx12 = _mm_sub_ps(ix1,jx2);
221 dy12 = _mm_sub_ps(iy1,jy2);
222 dz12 = _mm_sub_ps(iz1,jz2);
223 dx20 = _mm_sub_ps(ix2,jx0);
224 dy20 = _mm_sub_ps(iy2,jy0);
225 dz20 = _mm_sub_ps(iz2,jz0);
226 dx21 = _mm_sub_ps(ix2,jx1);
227 dy21 = _mm_sub_ps(iy2,jy1);
228 dz21 = _mm_sub_ps(iz2,jz1);
229 dx22 = _mm_sub_ps(ix2,jx2);
230 dy22 = _mm_sub_ps(iy2,jy2);
231 dz22 = _mm_sub_ps(iz2,jz2);
233 /* Calculate squared distance and things based on it */
234 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
235 rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
236 rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
237 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
238 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
239 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
240 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
241 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
242 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
244 rinv00 = gmx_mm_invsqrt_ps(rsq00);
245 rinv01 = gmx_mm_invsqrt_ps(rsq01);
246 rinv02 = gmx_mm_invsqrt_ps(rsq02);
247 rinv10 = gmx_mm_invsqrt_ps(rsq10);
248 rinv11 = gmx_mm_invsqrt_ps(rsq11);
249 rinv12 = gmx_mm_invsqrt_ps(rsq12);
250 rinv20 = gmx_mm_invsqrt_ps(rsq20);
251 rinv21 = gmx_mm_invsqrt_ps(rsq21);
252 rinv22 = gmx_mm_invsqrt_ps(rsq22);
254 fjx0 = _mm_setzero_ps();
255 fjy0 = _mm_setzero_ps();
256 fjz0 = _mm_setzero_ps();
257 fjx1 = _mm_setzero_ps();
258 fjy1 = _mm_setzero_ps();
259 fjz1 = _mm_setzero_ps();
260 fjx2 = _mm_setzero_ps();
261 fjy2 = _mm_setzero_ps();
262 fjz2 = _mm_setzero_ps();
264 /**************************
265 * CALCULATE INTERACTIONS *
266 **************************/
268 r00 = _mm_mul_ps(rsq00,rinv00);
270 /* Calculate table index by multiplying r with table scale and truncate to integer */
271 rt = _mm_mul_ps(r00,vftabscale);
272 vfitab = _mm_cvttps_epi32(rt);
273 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
274 vfitab = _mm_slli_epi32(vfitab,2);
276 /* CUBIC SPLINE TABLE ELECTROSTATICS */
277 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
278 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
279 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
280 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
281 _MM_TRANSPOSE4_PS(Y,F,G,H);
282 Heps = _mm_mul_ps(vfeps,H);
283 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
284 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
285 velec = _mm_mul_ps(qq00,VV);
286 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
287 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
289 /* Update potential sum for this i atom from the interaction with this j atom. */
290 velecsum = _mm_add_ps(velecsum,velec);
294 /* Calculate temporary vectorial force */
295 tx = _mm_mul_ps(fscal,dx00);
296 ty = _mm_mul_ps(fscal,dy00);
297 tz = _mm_mul_ps(fscal,dz00);
299 /* Update vectorial force */
300 fix0 = _mm_add_ps(fix0,tx);
301 fiy0 = _mm_add_ps(fiy0,ty);
302 fiz0 = _mm_add_ps(fiz0,tz);
304 fjx0 = _mm_add_ps(fjx0,tx);
305 fjy0 = _mm_add_ps(fjy0,ty);
306 fjz0 = _mm_add_ps(fjz0,tz);
308 /**************************
309 * CALCULATE INTERACTIONS *
310 **************************/
312 r01 = _mm_mul_ps(rsq01,rinv01);
314 /* Calculate table index by multiplying r with table scale and truncate to integer */
315 rt = _mm_mul_ps(r01,vftabscale);
316 vfitab = _mm_cvttps_epi32(rt);
317 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
318 vfitab = _mm_slli_epi32(vfitab,2);
320 /* CUBIC SPLINE TABLE ELECTROSTATICS */
321 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
322 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
323 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
324 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
325 _MM_TRANSPOSE4_PS(Y,F,G,H);
326 Heps = _mm_mul_ps(vfeps,H);
327 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
328 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
329 velec = _mm_mul_ps(qq01,VV);
330 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
331 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq01,FF),_mm_mul_ps(vftabscale,rinv01)));
333 /* Update potential sum for this i atom from the interaction with this j atom. */
334 velecsum = _mm_add_ps(velecsum,velec);
338 /* Calculate temporary vectorial force */
339 tx = _mm_mul_ps(fscal,dx01);
340 ty = _mm_mul_ps(fscal,dy01);
341 tz = _mm_mul_ps(fscal,dz01);
343 /* Update vectorial force */
344 fix0 = _mm_add_ps(fix0,tx);
345 fiy0 = _mm_add_ps(fiy0,ty);
346 fiz0 = _mm_add_ps(fiz0,tz);
348 fjx1 = _mm_add_ps(fjx1,tx);
349 fjy1 = _mm_add_ps(fjy1,ty);
350 fjz1 = _mm_add_ps(fjz1,tz);
352 /**************************
353 * CALCULATE INTERACTIONS *
354 **************************/
356 r02 = _mm_mul_ps(rsq02,rinv02);
358 /* Calculate table index by multiplying r with table scale and truncate to integer */
359 rt = _mm_mul_ps(r02,vftabscale);
360 vfitab = _mm_cvttps_epi32(rt);
361 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
362 vfitab = _mm_slli_epi32(vfitab,2);
364 /* CUBIC SPLINE TABLE ELECTROSTATICS */
365 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
366 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
367 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
368 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
369 _MM_TRANSPOSE4_PS(Y,F,G,H);
370 Heps = _mm_mul_ps(vfeps,H);
371 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
372 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
373 velec = _mm_mul_ps(qq02,VV);
374 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
375 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq02,FF),_mm_mul_ps(vftabscale,rinv02)));
377 /* Update potential sum for this i atom from the interaction with this j atom. */
378 velecsum = _mm_add_ps(velecsum,velec);
382 /* Calculate temporary vectorial force */
383 tx = _mm_mul_ps(fscal,dx02);
384 ty = _mm_mul_ps(fscal,dy02);
385 tz = _mm_mul_ps(fscal,dz02);
387 /* Update vectorial force */
388 fix0 = _mm_add_ps(fix0,tx);
389 fiy0 = _mm_add_ps(fiy0,ty);
390 fiz0 = _mm_add_ps(fiz0,tz);
392 fjx2 = _mm_add_ps(fjx2,tx);
393 fjy2 = _mm_add_ps(fjy2,ty);
394 fjz2 = _mm_add_ps(fjz2,tz);
396 /**************************
397 * CALCULATE INTERACTIONS *
398 **************************/
400 r10 = _mm_mul_ps(rsq10,rinv10);
402 /* Calculate table index by multiplying r with table scale and truncate to integer */
403 rt = _mm_mul_ps(r10,vftabscale);
404 vfitab = _mm_cvttps_epi32(rt);
405 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
406 vfitab = _mm_slli_epi32(vfitab,2);
408 /* CUBIC SPLINE TABLE ELECTROSTATICS */
409 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
410 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
411 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
412 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
413 _MM_TRANSPOSE4_PS(Y,F,G,H);
414 Heps = _mm_mul_ps(vfeps,H);
415 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
416 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
417 velec = _mm_mul_ps(qq10,VV);
418 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
419 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
421 /* Update potential sum for this i atom from the interaction with this j atom. */
422 velecsum = _mm_add_ps(velecsum,velec);
426 /* Calculate temporary vectorial force */
427 tx = _mm_mul_ps(fscal,dx10);
428 ty = _mm_mul_ps(fscal,dy10);
429 tz = _mm_mul_ps(fscal,dz10);
431 /* Update vectorial force */
432 fix1 = _mm_add_ps(fix1,tx);
433 fiy1 = _mm_add_ps(fiy1,ty);
434 fiz1 = _mm_add_ps(fiz1,tz);
436 fjx0 = _mm_add_ps(fjx0,tx);
437 fjy0 = _mm_add_ps(fjy0,ty);
438 fjz0 = _mm_add_ps(fjz0,tz);
440 /**************************
441 * CALCULATE INTERACTIONS *
442 **************************/
444 r11 = _mm_mul_ps(rsq11,rinv11);
446 /* Calculate table index by multiplying r with table scale and truncate to integer */
447 rt = _mm_mul_ps(r11,vftabscale);
448 vfitab = _mm_cvttps_epi32(rt);
449 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
450 vfitab = _mm_slli_epi32(vfitab,2);
452 /* CUBIC SPLINE TABLE ELECTROSTATICS */
453 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
454 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
455 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
456 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
457 _MM_TRANSPOSE4_PS(Y,F,G,H);
458 Heps = _mm_mul_ps(vfeps,H);
459 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
460 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
461 velec = _mm_mul_ps(qq11,VV);
462 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
463 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
465 /* Update potential sum for this i atom from the interaction with this j atom. */
466 velecsum = _mm_add_ps(velecsum,velec);
470 /* Calculate temporary vectorial force */
471 tx = _mm_mul_ps(fscal,dx11);
472 ty = _mm_mul_ps(fscal,dy11);
473 tz = _mm_mul_ps(fscal,dz11);
475 /* Update vectorial force */
476 fix1 = _mm_add_ps(fix1,tx);
477 fiy1 = _mm_add_ps(fiy1,ty);
478 fiz1 = _mm_add_ps(fiz1,tz);
480 fjx1 = _mm_add_ps(fjx1,tx);
481 fjy1 = _mm_add_ps(fjy1,ty);
482 fjz1 = _mm_add_ps(fjz1,tz);
484 /**************************
485 * CALCULATE INTERACTIONS *
486 **************************/
488 r12 = _mm_mul_ps(rsq12,rinv12);
490 /* Calculate table index by multiplying r with table scale and truncate to integer */
491 rt = _mm_mul_ps(r12,vftabscale);
492 vfitab = _mm_cvttps_epi32(rt);
493 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
494 vfitab = _mm_slli_epi32(vfitab,2);
496 /* CUBIC SPLINE TABLE ELECTROSTATICS */
497 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
498 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
499 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
500 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
501 _MM_TRANSPOSE4_PS(Y,F,G,H);
502 Heps = _mm_mul_ps(vfeps,H);
503 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
504 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
505 velec = _mm_mul_ps(qq12,VV);
506 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
507 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
509 /* Update potential sum for this i atom from the interaction with this j atom. */
510 velecsum = _mm_add_ps(velecsum,velec);
514 /* Calculate temporary vectorial force */
515 tx = _mm_mul_ps(fscal,dx12);
516 ty = _mm_mul_ps(fscal,dy12);
517 tz = _mm_mul_ps(fscal,dz12);
519 /* Update vectorial force */
520 fix1 = _mm_add_ps(fix1,tx);
521 fiy1 = _mm_add_ps(fiy1,ty);
522 fiz1 = _mm_add_ps(fiz1,tz);
524 fjx2 = _mm_add_ps(fjx2,tx);
525 fjy2 = _mm_add_ps(fjy2,ty);
526 fjz2 = _mm_add_ps(fjz2,tz);
528 /**************************
529 * CALCULATE INTERACTIONS *
530 **************************/
532 r20 = _mm_mul_ps(rsq20,rinv20);
534 /* Calculate table index by multiplying r with table scale and truncate to integer */
535 rt = _mm_mul_ps(r20,vftabscale);
536 vfitab = _mm_cvttps_epi32(rt);
537 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
538 vfitab = _mm_slli_epi32(vfitab,2);
540 /* CUBIC SPLINE TABLE ELECTROSTATICS */
541 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
542 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
543 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
544 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
545 _MM_TRANSPOSE4_PS(Y,F,G,H);
546 Heps = _mm_mul_ps(vfeps,H);
547 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
548 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
549 velec = _mm_mul_ps(qq20,VV);
550 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
551 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
553 /* Update potential sum for this i atom from the interaction with this j atom. */
554 velecsum = _mm_add_ps(velecsum,velec);
558 /* Calculate temporary vectorial force */
559 tx = _mm_mul_ps(fscal,dx20);
560 ty = _mm_mul_ps(fscal,dy20);
561 tz = _mm_mul_ps(fscal,dz20);
563 /* Update vectorial force */
564 fix2 = _mm_add_ps(fix2,tx);
565 fiy2 = _mm_add_ps(fiy2,ty);
566 fiz2 = _mm_add_ps(fiz2,tz);
568 fjx0 = _mm_add_ps(fjx0,tx);
569 fjy0 = _mm_add_ps(fjy0,ty);
570 fjz0 = _mm_add_ps(fjz0,tz);
572 /**************************
573 * CALCULATE INTERACTIONS *
574 **************************/
576 r21 = _mm_mul_ps(rsq21,rinv21);
578 /* Calculate table index by multiplying r with table scale and truncate to integer */
579 rt = _mm_mul_ps(r21,vftabscale);
580 vfitab = _mm_cvttps_epi32(rt);
581 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
582 vfitab = _mm_slli_epi32(vfitab,2);
584 /* CUBIC SPLINE TABLE ELECTROSTATICS */
585 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
586 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
587 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
588 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
589 _MM_TRANSPOSE4_PS(Y,F,G,H);
590 Heps = _mm_mul_ps(vfeps,H);
591 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
592 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
593 velec = _mm_mul_ps(qq21,VV);
594 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
595 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
597 /* Update potential sum for this i atom from the interaction with this j atom. */
598 velecsum = _mm_add_ps(velecsum,velec);
602 /* Calculate temporary vectorial force */
603 tx = _mm_mul_ps(fscal,dx21);
604 ty = _mm_mul_ps(fscal,dy21);
605 tz = _mm_mul_ps(fscal,dz21);
607 /* Update vectorial force */
608 fix2 = _mm_add_ps(fix2,tx);
609 fiy2 = _mm_add_ps(fiy2,ty);
610 fiz2 = _mm_add_ps(fiz2,tz);
612 fjx1 = _mm_add_ps(fjx1,tx);
613 fjy1 = _mm_add_ps(fjy1,ty);
614 fjz1 = _mm_add_ps(fjz1,tz);
616 /**************************
617 * CALCULATE INTERACTIONS *
618 **************************/
620 r22 = _mm_mul_ps(rsq22,rinv22);
622 /* Calculate table index by multiplying r with table scale and truncate to integer */
623 rt = _mm_mul_ps(r22,vftabscale);
624 vfitab = _mm_cvttps_epi32(rt);
625 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
626 vfitab = _mm_slli_epi32(vfitab,2);
628 /* CUBIC SPLINE TABLE ELECTROSTATICS */
629 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
630 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
631 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
632 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
633 _MM_TRANSPOSE4_PS(Y,F,G,H);
634 Heps = _mm_mul_ps(vfeps,H);
635 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
636 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
637 velec = _mm_mul_ps(qq22,VV);
638 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
639 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
641 /* Update potential sum for this i atom from the interaction with this j atom. */
642 velecsum = _mm_add_ps(velecsum,velec);
646 /* Calculate temporary vectorial force */
647 tx = _mm_mul_ps(fscal,dx22);
648 ty = _mm_mul_ps(fscal,dy22);
649 tz = _mm_mul_ps(fscal,dz22);
651 /* Update vectorial force */
652 fix2 = _mm_add_ps(fix2,tx);
653 fiy2 = _mm_add_ps(fiy2,ty);
654 fiz2 = _mm_add_ps(fiz2,tz);
656 fjx2 = _mm_add_ps(fjx2,tx);
657 fjy2 = _mm_add_ps(fjy2,ty);
658 fjz2 = _mm_add_ps(fjz2,tz);
660 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
661 f+j_coord_offsetC,f+j_coord_offsetD,
662 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
664 /* Inner loop uses 387 flops */
670 /* Get j neighbor index, and coordinate index */
676 /* Sign of each element will be negative for non-real atoms.
677 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
678 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
680 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
681 jnrA = (jnrA>=0) ? jnrA : 0;
682 jnrB = (jnrB>=0) ? jnrB : 0;
683 jnrC = (jnrC>=0) ? jnrC : 0;
684 jnrD = (jnrD>=0) ? jnrD : 0;
686 j_coord_offsetA = DIM*jnrA;
687 j_coord_offsetB = DIM*jnrB;
688 j_coord_offsetC = DIM*jnrC;
689 j_coord_offsetD = DIM*jnrD;
691 /* load j atom coordinates */
692 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
693 x+j_coord_offsetC,x+j_coord_offsetD,
694 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
696 /* Calculate displacement vector */
697 dx00 = _mm_sub_ps(ix0,jx0);
698 dy00 = _mm_sub_ps(iy0,jy0);
699 dz00 = _mm_sub_ps(iz0,jz0);
700 dx01 = _mm_sub_ps(ix0,jx1);
701 dy01 = _mm_sub_ps(iy0,jy1);
702 dz01 = _mm_sub_ps(iz0,jz1);
703 dx02 = _mm_sub_ps(ix0,jx2);
704 dy02 = _mm_sub_ps(iy0,jy2);
705 dz02 = _mm_sub_ps(iz0,jz2);
706 dx10 = _mm_sub_ps(ix1,jx0);
707 dy10 = _mm_sub_ps(iy1,jy0);
708 dz10 = _mm_sub_ps(iz1,jz0);
709 dx11 = _mm_sub_ps(ix1,jx1);
710 dy11 = _mm_sub_ps(iy1,jy1);
711 dz11 = _mm_sub_ps(iz1,jz1);
712 dx12 = _mm_sub_ps(ix1,jx2);
713 dy12 = _mm_sub_ps(iy1,jy2);
714 dz12 = _mm_sub_ps(iz1,jz2);
715 dx20 = _mm_sub_ps(ix2,jx0);
716 dy20 = _mm_sub_ps(iy2,jy0);
717 dz20 = _mm_sub_ps(iz2,jz0);
718 dx21 = _mm_sub_ps(ix2,jx1);
719 dy21 = _mm_sub_ps(iy2,jy1);
720 dz21 = _mm_sub_ps(iz2,jz1);
721 dx22 = _mm_sub_ps(ix2,jx2);
722 dy22 = _mm_sub_ps(iy2,jy2);
723 dz22 = _mm_sub_ps(iz2,jz2);
725 /* Calculate squared distance and things based on it */
726 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
727 rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
728 rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
729 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
730 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
731 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
732 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
733 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
734 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
736 rinv00 = gmx_mm_invsqrt_ps(rsq00);
737 rinv01 = gmx_mm_invsqrt_ps(rsq01);
738 rinv02 = gmx_mm_invsqrt_ps(rsq02);
739 rinv10 = gmx_mm_invsqrt_ps(rsq10);
740 rinv11 = gmx_mm_invsqrt_ps(rsq11);
741 rinv12 = gmx_mm_invsqrt_ps(rsq12);
742 rinv20 = gmx_mm_invsqrt_ps(rsq20);
743 rinv21 = gmx_mm_invsqrt_ps(rsq21);
744 rinv22 = gmx_mm_invsqrt_ps(rsq22);
746 fjx0 = _mm_setzero_ps();
747 fjy0 = _mm_setzero_ps();
748 fjz0 = _mm_setzero_ps();
749 fjx1 = _mm_setzero_ps();
750 fjy1 = _mm_setzero_ps();
751 fjz1 = _mm_setzero_ps();
752 fjx2 = _mm_setzero_ps();
753 fjy2 = _mm_setzero_ps();
754 fjz2 = _mm_setzero_ps();
756 /**************************
757 * CALCULATE INTERACTIONS *
758 **************************/
760 r00 = _mm_mul_ps(rsq00,rinv00);
761 r00 = _mm_andnot_ps(dummy_mask,r00);
763 /* Calculate table index by multiplying r with table scale and truncate to integer */
764 rt = _mm_mul_ps(r00,vftabscale);
765 vfitab = _mm_cvttps_epi32(rt);
766 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
767 vfitab = _mm_slli_epi32(vfitab,2);
769 /* CUBIC SPLINE TABLE ELECTROSTATICS */
770 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
771 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
772 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
773 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
774 _MM_TRANSPOSE4_PS(Y,F,G,H);
775 Heps = _mm_mul_ps(vfeps,H);
776 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
777 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
778 velec = _mm_mul_ps(qq00,VV);
779 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
780 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
782 /* Update potential sum for this i atom from the interaction with this j atom. */
783 velec = _mm_andnot_ps(dummy_mask,velec);
784 velecsum = _mm_add_ps(velecsum,velec);
788 fscal = _mm_andnot_ps(dummy_mask,fscal);
790 /* Calculate temporary vectorial force */
791 tx = _mm_mul_ps(fscal,dx00);
792 ty = _mm_mul_ps(fscal,dy00);
793 tz = _mm_mul_ps(fscal,dz00);
795 /* Update vectorial force */
796 fix0 = _mm_add_ps(fix0,tx);
797 fiy0 = _mm_add_ps(fiy0,ty);
798 fiz0 = _mm_add_ps(fiz0,tz);
800 fjx0 = _mm_add_ps(fjx0,tx);
801 fjy0 = _mm_add_ps(fjy0,ty);
802 fjz0 = _mm_add_ps(fjz0,tz);
804 /**************************
805 * CALCULATE INTERACTIONS *
806 **************************/
808 r01 = _mm_mul_ps(rsq01,rinv01);
809 r01 = _mm_andnot_ps(dummy_mask,r01);
811 /* Calculate table index by multiplying r with table scale and truncate to integer */
812 rt = _mm_mul_ps(r01,vftabscale);
813 vfitab = _mm_cvttps_epi32(rt);
814 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
815 vfitab = _mm_slli_epi32(vfitab,2);
817 /* CUBIC SPLINE TABLE ELECTROSTATICS */
818 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
819 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
820 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
821 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
822 _MM_TRANSPOSE4_PS(Y,F,G,H);
823 Heps = _mm_mul_ps(vfeps,H);
824 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
825 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
826 velec = _mm_mul_ps(qq01,VV);
827 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
828 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq01,FF),_mm_mul_ps(vftabscale,rinv01)));
830 /* Update potential sum for this i atom from the interaction with this j atom. */
831 velec = _mm_andnot_ps(dummy_mask,velec);
832 velecsum = _mm_add_ps(velecsum,velec);
836 fscal = _mm_andnot_ps(dummy_mask,fscal);
838 /* Calculate temporary vectorial force */
839 tx = _mm_mul_ps(fscal,dx01);
840 ty = _mm_mul_ps(fscal,dy01);
841 tz = _mm_mul_ps(fscal,dz01);
843 /* Update vectorial force */
844 fix0 = _mm_add_ps(fix0,tx);
845 fiy0 = _mm_add_ps(fiy0,ty);
846 fiz0 = _mm_add_ps(fiz0,tz);
848 fjx1 = _mm_add_ps(fjx1,tx);
849 fjy1 = _mm_add_ps(fjy1,ty);
850 fjz1 = _mm_add_ps(fjz1,tz);
852 /**************************
853 * CALCULATE INTERACTIONS *
854 **************************/
856 r02 = _mm_mul_ps(rsq02,rinv02);
857 r02 = _mm_andnot_ps(dummy_mask,r02);
859 /* Calculate table index by multiplying r with table scale and truncate to integer */
860 rt = _mm_mul_ps(r02,vftabscale);
861 vfitab = _mm_cvttps_epi32(rt);
862 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
863 vfitab = _mm_slli_epi32(vfitab,2);
865 /* CUBIC SPLINE TABLE ELECTROSTATICS */
866 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
867 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
868 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
869 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
870 _MM_TRANSPOSE4_PS(Y,F,G,H);
871 Heps = _mm_mul_ps(vfeps,H);
872 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
873 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
874 velec = _mm_mul_ps(qq02,VV);
875 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
876 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq02,FF),_mm_mul_ps(vftabscale,rinv02)));
878 /* Update potential sum for this i atom from the interaction with this j atom. */
879 velec = _mm_andnot_ps(dummy_mask,velec);
880 velecsum = _mm_add_ps(velecsum,velec);
884 fscal = _mm_andnot_ps(dummy_mask,fscal);
886 /* Calculate temporary vectorial force */
887 tx = _mm_mul_ps(fscal,dx02);
888 ty = _mm_mul_ps(fscal,dy02);
889 tz = _mm_mul_ps(fscal,dz02);
891 /* Update vectorial force */
892 fix0 = _mm_add_ps(fix0,tx);
893 fiy0 = _mm_add_ps(fiy0,ty);
894 fiz0 = _mm_add_ps(fiz0,tz);
896 fjx2 = _mm_add_ps(fjx2,tx);
897 fjy2 = _mm_add_ps(fjy2,ty);
898 fjz2 = _mm_add_ps(fjz2,tz);
900 /**************************
901 * CALCULATE INTERACTIONS *
902 **************************/
904 r10 = _mm_mul_ps(rsq10,rinv10);
905 r10 = _mm_andnot_ps(dummy_mask,r10);
907 /* Calculate table index by multiplying r with table scale and truncate to integer */
908 rt = _mm_mul_ps(r10,vftabscale);
909 vfitab = _mm_cvttps_epi32(rt);
910 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
911 vfitab = _mm_slli_epi32(vfitab,2);
913 /* CUBIC SPLINE TABLE ELECTROSTATICS */
914 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
915 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
916 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
917 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
918 _MM_TRANSPOSE4_PS(Y,F,G,H);
919 Heps = _mm_mul_ps(vfeps,H);
920 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
921 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
922 velec = _mm_mul_ps(qq10,VV);
923 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
924 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
926 /* Update potential sum for this i atom from the interaction with this j atom. */
927 velec = _mm_andnot_ps(dummy_mask,velec);
928 velecsum = _mm_add_ps(velecsum,velec);
932 fscal = _mm_andnot_ps(dummy_mask,fscal);
934 /* Calculate temporary vectorial force */
935 tx = _mm_mul_ps(fscal,dx10);
936 ty = _mm_mul_ps(fscal,dy10);
937 tz = _mm_mul_ps(fscal,dz10);
939 /* Update vectorial force */
940 fix1 = _mm_add_ps(fix1,tx);
941 fiy1 = _mm_add_ps(fiy1,ty);
942 fiz1 = _mm_add_ps(fiz1,tz);
944 fjx0 = _mm_add_ps(fjx0,tx);
945 fjy0 = _mm_add_ps(fjy0,ty);
946 fjz0 = _mm_add_ps(fjz0,tz);
948 /**************************
949 * CALCULATE INTERACTIONS *
950 **************************/
952 r11 = _mm_mul_ps(rsq11,rinv11);
953 r11 = _mm_andnot_ps(dummy_mask,r11);
955 /* Calculate table index by multiplying r with table scale and truncate to integer */
956 rt = _mm_mul_ps(r11,vftabscale);
957 vfitab = _mm_cvttps_epi32(rt);
958 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
959 vfitab = _mm_slli_epi32(vfitab,2);
961 /* CUBIC SPLINE TABLE ELECTROSTATICS */
962 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
963 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
964 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
965 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
966 _MM_TRANSPOSE4_PS(Y,F,G,H);
967 Heps = _mm_mul_ps(vfeps,H);
968 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
969 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
970 velec = _mm_mul_ps(qq11,VV);
971 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
972 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
974 /* Update potential sum for this i atom from the interaction with this j atom. */
975 velec = _mm_andnot_ps(dummy_mask,velec);
976 velecsum = _mm_add_ps(velecsum,velec);
980 fscal = _mm_andnot_ps(dummy_mask,fscal);
982 /* Calculate temporary vectorial force */
983 tx = _mm_mul_ps(fscal,dx11);
984 ty = _mm_mul_ps(fscal,dy11);
985 tz = _mm_mul_ps(fscal,dz11);
987 /* Update vectorial force */
988 fix1 = _mm_add_ps(fix1,tx);
989 fiy1 = _mm_add_ps(fiy1,ty);
990 fiz1 = _mm_add_ps(fiz1,tz);
992 fjx1 = _mm_add_ps(fjx1,tx);
993 fjy1 = _mm_add_ps(fjy1,ty);
994 fjz1 = _mm_add_ps(fjz1,tz);
996 /**************************
997 * CALCULATE INTERACTIONS *
998 **************************/
1000 r12 = _mm_mul_ps(rsq12,rinv12);
1001 r12 = _mm_andnot_ps(dummy_mask,r12);
1003 /* Calculate table index by multiplying r with table scale and truncate to integer */
1004 rt = _mm_mul_ps(r12,vftabscale);
1005 vfitab = _mm_cvttps_epi32(rt);
1006 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1007 vfitab = _mm_slli_epi32(vfitab,2);
1009 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1010 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1011 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1012 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1013 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1014 _MM_TRANSPOSE4_PS(Y,F,G,H);
1015 Heps = _mm_mul_ps(vfeps,H);
1016 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1017 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
1018 velec = _mm_mul_ps(qq12,VV);
1019 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1020 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
1022 /* Update potential sum for this i atom from the interaction with this j atom. */
1023 velec = _mm_andnot_ps(dummy_mask,velec);
1024 velecsum = _mm_add_ps(velecsum,velec);
1028 fscal = _mm_andnot_ps(dummy_mask,fscal);
1030 /* Calculate temporary vectorial force */
1031 tx = _mm_mul_ps(fscal,dx12);
1032 ty = _mm_mul_ps(fscal,dy12);
1033 tz = _mm_mul_ps(fscal,dz12);
1035 /* Update vectorial force */
1036 fix1 = _mm_add_ps(fix1,tx);
1037 fiy1 = _mm_add_ps(fiy1,ty);
1038 fiz1 = _mm_add_ps(fiz1,tz);
1040 fjx2 = _mm_add_ps(fjx2,tx);
1041 fjy2 = _mm_add_ps(fjy2,ty);
1042 fjz2 = _mm_add_ps(fjz2,tz);
1044 /**************************
1045 * CALCULATE INTERACTIONS *
1046 **************************/
1048 r20 = _mm_mul_ps(rsq20,rinv20);
1049 r20 = _mm_andnot_ps(dummy_mask,r20);
1051 /* Calculate table index by multiplying r with table scale and truncate to integer */
1052 rt = _mm_mul_ps(r20,vftabscale);
1053 vfitab = _mm_cvttps_epi32(rt);
1054 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1055 vfitab = _mm_slli_epi32(vfitab,2);
1057 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1058 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1059 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1060 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1061 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1062 _MM_TRANSPOSE4_PS(Y,F,G,H);
1063 Heps = _mm_mul_ps(vfeps,H);
1064 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1065 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
1066 velec = _mm_mul_ps(qq20,VV);
1067 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1068 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1070 /* Update potential sum for this i atom from the interaction with this j atom. */
1071 velec = _mm_andnot_ps(dummy_mask,velec);
1072 velecsum = _mm_add_ps(velecsum,velec);
1076 fscal = _mm_andnot_ps(dummy_mask,fscal);
1078 /* Calculate temporary vectorial force */
1079 tx = _mm_mul_ps(fscal,dx20);
1080 ty = _mm_mul_ps(fscal,dy20);
1081 tz = _mm_mul_ps(fscal,dz20);
1083 /* Update vectorial force */
1084 fix2 = _mm_add_ps(fix2,tx);
1085 fiy2 = _mm_add_ps(fiy2,ty);
1086 fiz2 = _mm_add_ps(fiz2,tz);
1088 fjx0 = _mm_add_ps(fjx0,tx);
1089 fjy0 = _mm_add_ps(fjy0,ty);
1090 fjz0 = _mm_add_ps(fjz0,tz);
1092 /**************************
1093 * CALCULATE INTERACTIONS *
1094 **************************/
1096 r21 = _mm_mul_ps(rsq21,rinv21);
1097 r21 = _mm_andnot_ps(dummy_mask,r21);
1099 /* Calculate table index by multiplying r with table scale and truncate to integer */
1100 rt = _mm_mul_ps(r21,vftabscale);
1101 vfitab = _mm_cvttps_epi32(rt);
1102 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1103 vfitab = _mm_slli_epi32(vfitab,2);
1105 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1106 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1107 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1108 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1109 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1110 _MM_TRANSPOSE4_PS(Y,F,G,H);
1111 Heps = _mm_mul_ps(vfeps,H);
1112 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1113 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
1114 velec = _mm_mul_ps(qq21,VV);
1115 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1116 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
1118 /* Update potential sum for this i atom from the interaction with this j atom. */
1119 velec = _mm_andnot_ps(dummy_mask,velec);
1120 velecsum = _mm_add_ps(velecsum,velec);
1124 fscal = _mm_andnot_ps(dummy_mask,fscal);
1126 /* Calculate temporary vectorial force */
1127 tx = _mm_mul_ps(fscal,dx21);
1128 ty = _mm_mul_ps(fscal,dy21);
1129 tz = _mm_mul_ps(fscal,dz21);
1131 /* Update vectorial force */
1132 fix2 = _mm_add_ps(fix2,tx);
1133 fiy2 = _mm_add_ps(fiy2,ty);
1134 fiz2 = _mm_add_ps(fiz2,tz);
1136 fjx1 = _mm_add_ps(fjx1,tx);
1137 fjy1 = _mm_add_ps(fjy1,ty);
1138 fjz1 = _mm_add_ps(fjz1,tz);
1140 /**************************
1141 * CALCULATE INTERACTIONS *
1142 **************************/
1144 r22 = _mm_mul_ps(rsq22,rinv22);
1145 r22 = _mm_andnot_ps(dummy_mask,r22);
1147 /* Calculate table index by multiplying r with table scale and truncate to integer */
1148 rt = _mm_mul_ps(r22,vftabscale);
1149 vfitab = _mm_cvttps_epi32(rt);
1150 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1151 vfitab = _mm_slli_epi32(vfitab,2);
1153 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1154 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1155 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1156 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1157 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1158 _MM_TRANSPOSE4_PS(Y,F,G,H);
1159 Heps = _mm_mul_ps(vfeps,H);
1160 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1161 VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
1162 velec = _mm_mul_ps(qq22,VV);
1163 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1164 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
1166 /* Update potential sum for this i atom from the interaction with this j atom. */
1167 velec = _mm_andnot_ps(dummy_mask,velec);
1168 velecsum = _mm_add_ps(velecsum,velec);
1172 fscal = _mm_andnot_ps(dummy_mask,fscal);
1174 /* Calculate temporary vectorial force */
1175 tx = _mm_mul_ps(fscal,dx22);
1176 ty = _mm_mul_ps(fscal,dy22);
1177 tz = _mm_mul_ps(fscal,dz22);
1179 /* Update vectorial force */
1180 fix2 = _mm_add_ps(fix2,tx);
1181 fiy2 = _mm_add_ps(fiy2,ty);
1182 fiz2 = _mm_add_ps(fiz2,tz);
1184 fjx2 = _mm_add_ps(fjx2,tx);
1185 fjy2 = _mm_add_ps(fjy2,ty);
1186 fjz2 = _mm_add_ps(fjz2,tz);
1188 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1189 f+j_coord_offsetC,f+j_coord_offsetD,
1190 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1192 /* Inner loop uses 396 flops */
1195 /* End of innermost loop */
1197 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1198 f+i_coord_offset,fshift+i_shift_offset);
1201 /* Update potential energies */
1202 gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1204 /* Increment number of inner iterations */
1205 inneriter += j_index_end - j_index_start;
1207 /* Outer loop uses 28 flops */
1210 /* Increment number of outer iterations */
1213 /* Update outer/inner flops */
1215 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_VF,outeriter*28 + inneriter*396);
1218 * Gromacs nonbonded kernel: nb_kernel_ElecCSTab_VdwNone_GeomW3W3_F_sse2_single
1219 * Electrostatics interaction: CubicSplineTable
1220 * VdW interaction: None
1221 * Geometry: Water3-Water3
1222 * Calculate force/pot: Force
1225 nb_kernel_ElecCSTab_VdwNone_GeomW3W3_F_sse2_single
1226 (t_nblist * gmx_restrict nlist,
1227 rvec * gmx_restrict xx,
1228 rvec * gmx_restrict ff,
1229 t_forcerec * gmx_restrict fr,
1230 t_mdatoms * gmx_restrict mdatoms,
1231 nb_kernel_data_t * gmx_restrict kernel_data,
1232 t_nrnb * gmx_restrict nrnb)
1234 /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1235 * just 0 for non-waters.
1236 * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
1237 * jnr indices corresponding to data put in the four positions in the SIMD register.
1239 int i_shift_offset,i_coord_offset,outeriter,inneriter;
1240 int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1241 int jnrA,jnrB,jnrC,jnrD;
1242 int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1243 int *iinr,*jindex,*jjnr,*shiftidx,*gid;
1244 real shX,shY,shZ,rcutoff_scalar;
1245 real *shiftvec,*fshift,*x,*f;
1246 __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1248 __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1250 __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1252 __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1253 int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1254 __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1255 int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1256 __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1257 int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1258 __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1259 __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1260 __m128 dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1261 __m128 dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1262 __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1263 __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1264 __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1265 __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1266 __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1267 __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1268 __m128 velec,felec,velecsum,facel,crf,krf,krf2;
1271 __m128i ifour = _mm_set1_epi32(4);
1272 __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
1274 __m128 dummy_mask,cutoff_mask;
1275 __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1276 __m128 one = _mm_set1_ps(1.0);
1277 __m128 two = _mm_set1_ps(2.0);
1283 jindex = nlist->jindex;
1285 shiftidx = nlist->shift;
1287 shiftvec = fr->shift_vec[0];
1288 fshift = fr->fshift[0];
1289 facel = _mm_set1_ps(fr->epsfac);
1290 charge = mdatoms->chargeA;
1292 vftab = kernel_data->table_elec->data;
1293 vftabscale = _mm_set1_ps(kernel_data->table_elec->scale);
1295 /* Setup water-specific parameters */
1296 inr = nlist->iinr[0];
1297 iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
1298 iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1299 iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1301 jq0 = _mm_set1_ps(charge[inr+0]);
1302 jq1 = _mm_set1_ps(charge[inr+1]);
1303 jq2 = _mm_set1_ps(charge[inr+2]);
1304 qq00 = _mm_mul_ps(iq0,jq0);
1305 qq01 = _mm_mul_ps(iq0,jq1);
1306 qq02 = _mm_mul_ps(iq0,jq2);
1307 qq10 = _mm_mul_ps(iq1,jq0);
1308 qq11 = _mm_mul_ps(iq1,jq1);
1309 qq12 = _mm_mul_ps(iq1,jq2);
1310 qq20 = _mm_mul_ps(iq2,jq0);
1311 qq21 = _mm_mul_ps(iq2,jq1);
1312 qq22 = _mm_mul_ps(iq2,jq2);
1314 /* Avoid stupid compiler warnings */
1315 jnrA = jnrB = jnrC = jnrD = 0;
1316 j_coord_offsetA = 0;
1317 j_coord_offsetB = 0;
1318 j_coord_offsetC = 0;
1319 j_coord_offsetD = 0;
1324 /* Start outer loop over neighborlists */
1325 for(iidx=0; iidx<nri; iidx++)
1327 /* Load shift vector for this list */
1328 i_shift_offset = DIM*shiftidx[iidx];
1329 shX = shiftvec[i_shift_offset+XX];
1330 shY = shiftvec[i_shift_offset+YY];
1331 shZ = shiftvec[i_shift_offset+ZZ];
1333 /* Load limits for loop over neighbors */
1334 j_index_start = jindex[iidx];
1335 j_index_end = jindex[iidx+1];
1337 /* Get outer coordinate index */
1339 i_coord_offset = DIM*inr;
1341 /* Load i particle coords and add shift vector */
1342 ix0 = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
1343 iy0 = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
1344 iz0 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
1345 ix1 = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
1346 iy1 = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
1347 iz1 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
1348 ix2 = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
1349 iy2 = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
1350 iz2 = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
1352 fix0 = _mm_setzero_ps();
1353 fiy0 = _mm_setzero_ps();
1354 fiz0 = _mm_setzero_ps();
1355 fix1 = _mm_setzero_ps();
1356 fiy1 = _mm_setzero_ps();
1357 fiz1 = _mm_setzero_ps();
1358 fix2 = _mm_setzero_ps();
1359 fiy2 = _mm_setzero_ps();
1360 fiz2 = _mm_setzero_ps();
1362 /* Start inner kernel loop */
1363 for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1366 /* Get j neighbor index, and coordinate index */
1368 jnrB = jjnr[jidx+1];
1369 jnrC = jjnr[jidx+2];
1370 jnrD = jjnr[jidx+3];
1372 j_coord_offsetA = DIM*jnrA;
1373 j_coord_offsetB = DIM*jnrB;
1374 j_coord_offsetC = DIM*jnrC;
1375 j_coord_offsetD = DIM*jnrD;
1377 /* load j atom coordinates */
1378 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1379 x+j_coord_offsetC,x+j_coord_offsetD,
1380 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1382 /* Calculate displacement vector */
1383 dx00 = _mm_sub_ps(ix0,jx0);
1384 dy00 = _mm_sub_ps(iy0,jy0);
1385 dz00 = _mm_sub_ps(iz0,jz0);
1386 dx01 = _mm_sub_ps(ix0,jx1);
1387 dy01 = _mm_sub_ps(iy0,jy1);
1388 dz01 = _mm_sub_ps(iz0,jz1);
1389 dx02 = _mm_sub_ps(ix0,jx2);
1390 dy02 = _mm_sub_ps(iy0,jy2);
1391 dz02 = _mm_sub_ps(iz0,jz2);
1392 dx10 = _mm_sub_ps(ix1,jx0);
1393 dy10 = _mm_sub_ps(iy1,jy0);
1394 dz10 = _mm_sub_ps(iz1,jz0);
1395 dx11 = _mm_sub_ps(ix1,jx1);
1396 dy11 = _mm_sub_ps(iy1,jy1);
1397 dz11 = _mm_sub_ps(iz1,jz1);
1398 dx12 = _mm_sub_ps(ix1,jx2);
1399 dy12 = _mm_sub_ps(iy1,jy2);
1400 dz12 = _mm_sub_ps(iz1,jz2);
1401 dx20 = _mm_sub_ps(ix2,jx0);
1402 dy20 = _mm_sub_ps(iy2,jy0);
1403 dz20 = _mm_sub_ps(iz2,jz0);
1404 dx21 = _mm_sub_ps(ix2,jx1);
1405 dy21 = _mm_sub_ps(iy2,jy1);
1406 dz21 = _mm_sub_ps(iz2,jz1);
1407 dx22 = _mm_sub_ps(ix2,jx2);
1408 dy22 = _mm_sub_ps(iy2,jy2);
1409 dz22 = _mm_sub_ps(iz2,jz2);
1411 /* Calculate squared distance and things based on it */
1412 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1413 rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1414 rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1415 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1416 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1417 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1418 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1419 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1420 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1422 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1423 rinv01 = gmx_mm_invsqrt_ps(rsq01);
1424 rinv02 = gmx_mm_invsqrt_ps(rsq02);
1425 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1426 rinv11 = gmx_mm_invsqrt_ps(rsq11);
1427 rinv12 = gmx_mm_invsqrt_ps(rsq12);
1428 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1429 rinv21 = gmx_mm_invsqrt_ps(rsq21);
1430 rinv22 = gmx_mm_invsqrt_ps(rsq22);
1432 fjx0 = _mm_setzero_ps();
1433 fjy0 = _mm_setzero_ps();
1434 fjz0 = _mm_setzero_ps();
1435 fjx1 = _mm_setzero_ps();
1436 fjy1 = _mm_setzero_ps();
1437 fjz1 = _mm_setzero_ps();
1438 fjx2 = _mm_setzero_ps();
1439 fjy2 = _mm_setzero_ps();
1440 fjz2 = _mm_setzero_ps();
1442 /**************************
1443 * CALCULATE INTERACTIONS *
1444 **************************/
1446 r00 = _mm_mul_ps(rsq00,rinv00);
1448 /* Calculate table index by multiplying r with table scale and truncate to integer */
1449 rt = _mm_mul_ps(r00,vftabscale);
1450 vfitab = _mm_cvttps_epi32(rt);
1451 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1452 vfitab = _mm_slli_epi32(vfitab,2);
1454 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1455 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1456 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1457 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1458 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1459 _MM_TRANSPOSE4_PS(Y,F,G,H);
1460 Heps = _mm_mul_ps(vfeps,H);
1461 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1462 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1463 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1467 /* Calculate temporary vectorial force */
1468 tx = _mm_mul_ps(fscal,dx00);
1469 ty = _mm_mul_ps(fscal,dy00);
1470 tz = _mm_mul_ps(fscal,dz00);
1472 /* Update vectorial force */
1473 fix0 = _mm_add_ps(fix0,tx);
1474 fiy0 = _mm_add_ps(fiy0,ty);
1475 fiz0 = _mm_add_ps(fiz0,tz);
1477 fjx0 = _mm_add_ps(fjx0,tx);
1478 fjy0 = _mm_add_ps(fjy0,ty);
1479 fjz0 = _mm_add_ps(fjz0,tz);
1481 /**************************
1482 * CALCULATE INTERACTIONS *
1483 **************************/
1485 r01 = _mm_mul_ps(rsq01,rinv01);
1487 /* Calculate table index by multiplying r with table scale and truncate to integer */
1488 rt = _mm_mul_ps(r01,vftabscale);
1489 vfitab = _mm_cvttps_epi32(rt);
1490 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1491 vfitab = _mm_slli_epi32(vfitab,2);
1493 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1494 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1495 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1496 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1497 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1498 _MM_TRANSPOSE4_PS(Y,F,G,H);
1499 Heps = _mm_mul_ps(vfeps,H);
1500 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1501 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1502 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq01,FF),_mm_mul_ps(vftabscale,rinv01)));
1506 /* Calculate temporary vectorial force */
1507 tx = _mm_mul_ps(fscal,dx01);
1508 ty = _mm_mul_ps(fscal,dy01);
1509 tz = _mm_mul_ps(fscal,dz01);
1511 /* Update vectorial force */
1512 fix0 = _mm_add_ps(fix0,tx);
1513 fiy0 = _mm_add_ps(fiy0,ty);
1514 fiz0 = _mm_add_ps(fiz0,tz);
1516 fjx1 = _mm_add_ps(fjx1,tx);
1517 fjy1 = _mm_add_ps(fjy1,ty);
1518 fjz1 = _mm_add_ps(fjz1,tz);
1520 /**************************
1521 * CALCULATE INTERACTIONS *
1522 **************************/
1524 r02 = _mm_mul_ps(rsq02,rinv02);
1526 /* Calculate table index by multiplying r with table scale and truncate to integer */
1527 rt = _mm_mul_ps(r02,vftabscale);
1528 vfitab = _mm_cvttps_epi32(rt);
1529 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1530 vfitab = _mm_slli_epi32(vfitab,2);
1532 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1533 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1534 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1535 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1536 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1537 _MM_TRANSPOSE4_PS(Y,F,G,H);
1538 Heps = _mm_mul_ps(vfeps,H);
1539 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1540 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1541 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq02,FF),_mm_mul_ps(vftabscale,rinv02)));
1545 /* Calculate temporary vectorial force */
1546 tx = _mm_mul_ps(fscal,dx02);
1547 ty = _mm_mul_ps(fscal,dy02);
1548 tz = _mm_mul_ps(fscal,dz02);
1550 /* Update vectorial force */
1551 fix0 = _mm_add_ps(fix0,tx);
1552 fiy0 = _mm_add_ps(fiy0,ty);
1553 fiz0 = _mm_add_ps(fiz0,tz);
1555 fjx2 = _mm_add_ps(fjx2,tx);
1556 fjy2 = _mm_add_ps(fjy2,ty);
1557 fjz2 = _mm_add_ps(fjz2,tz);
1559 /**************************
1560 * CALCULATE INTERACTIONS *
1561 **************************/
1563 r10 = _mm_mul_ps(rsq10,rinv10);
1565 /* Calculate table index by multiplying r with table scale and truncate to integer */
1566 rt = _mm_mul_ps(r10,vftabscale);
1567 vfitab = _mm_cvttps_epi32(rt);
1568 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1569 vfitab = _mm_slli_epi32(vfitab,2);
1571 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1572 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1573 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1574 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1575 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1576 _MM_TRANSPOSE4_PS(Y,F,G,H);
1577 Heps = _mm_mul_ps(vfeps,H);
1578 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1579 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1580 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1584 /* Calculate temporary vectorial force */
1585 tx = _mm_mul_ps(fscal,dx10);
1586 ty = _mm_mul_ps(fscal,dy10);
1587 tz = _mm_mul_ps(fscal,dz10);
1589 /* Update vectorial force */
1590 fix1 = _mm_add_ps(fix1,tx);
1591 fiy1 = _mm_add_ps(fiy1,ty);
1592 fiz1 = _mm_add_ps(fiz1,tz);
1594 fjx0 = _mm_add_ps(fjx0,tx);
1595 fjy0 = _mm_add_ps(fjy0,ty);
1596 fjz0 = _mm_add_ps(fjz0,tz);
1598 /**************************
1599 * CALCULATE INTERACTIONS *
1600 **************************/
1602 r11 = _mm_mul_ps(rsq11,rinv11);
1604 /* Calculate table index by multiplying r with table scale and truncate to integer */
1605 rt = _mm_mul_ps(r11,vftabscale);
1606 vfitab = _mm_cvttps_epi32(rt);
1607 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1608 vfitab = _mm_slli_epi32(vfitab,2);
1610 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1611 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1612 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1613 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1614 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1615 _MM_TRANSPOSE4_PS(Y,F,G,H);
1616 Heps = _mm_mul_ps(vfeps,H);
1617 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1618 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1619 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
1623 /* Calculate temporary vectorial force */
1624 tx = _mm_mul_ps(fscal,dx11);
1625 ty = _mm_mul_ps(fscal,dy11);
1626 tz = _mm_mul_ps(fscal,dz11);
1628 /* Update vectorial force */
1629 fix1 = _mm_add_ps(fix1,tx);
1630 fiy1 = _mm_add_ps(fiy1,ty);
1631 fiz1 = _mm_add_ps(fiz1,tz);
1633 fjx1 = _mm_add_ps(fjx1,tx);
1634 fjy1 = _mm_add_ps(fjy1,ty);
1635 fjz1 = _mm_add_ps(fjz1,tz);
1637 /**************************
1638 * CALCULATE INTERACTIONS *
1639 **************************/
1641 r12 = _mm_mul_ps(rsq12,rinv12);
1643 /* Calculate table index by multiplying r with table scale and truncate to integer */
1644 rt = _mm_mul_ps(r12,vftabscale);
1645 vfitab = _mm_cvttps_epi32(rt);
1646 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1647 vfitab = _mm_slli_epi32(vfitab,2);
1649 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1650 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1651 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1652 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1653 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1654 _MM_TRANSPOSE4_PS(Y,F,G,H);
1655 Heps = _mm_mul_ps(vfeps,H);
1656 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1657 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1658 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
1662 /* Calculate temporary vectorial force */
1663 tx = _mm_mul_ps(fscal,dx12);
1664 ty = _mm_mul_ps(fscal,dy12);
1665 tz = _mm_mul_ps(fscal,dz12);
1667 /* Update vectorial force */
1668 fix1 = _mm_add_ps(fix1,tx);
1669 fiy1 = _mm_add_ps(fiy1,ty);
1670 fiz1 = _mm_add_ps(fiz1,tz);
1672 fjx2 = _mm_add_ps(fjx2,tx);
1673 fjy2 = _mm_add_ps(fjy2,ty);
1674 fjz2 = _mm_add_ps(fjz2,tz);
1676 /**************************
1677 * CALCULATE INTERACTIONS *
1678 **************************/
1680 r20 = _mm_mul_ps(rsq20,rinv20);
1682 /* Calculate table index by multiplying r with table scale and truncate to integer */
1683 rt = _mm_mul_ps(r20,vftabscale);
1684 vfitab = _mm_cvttps_epi32(rt);
1685 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1686 vfitab = _mm_slli_epi32(vfitab,2);
1688 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1689 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1690 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1691 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1692 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1693 _MM_TRANSPOSE4_PS(Y,F,G,H);
1694 Heps = _mm_mul_ps(vfeps,H);
1695 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1696 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1697 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1701 /* Calculate temporary vectorial force */
1702 tx = _mm_mul_ps(fscal,dx20);
1703 ty = _mm_mul_ps(fscal,dy20);
1704 tz = _mm_mul_ps(fscal,dz20);
1706 /* Update vectorial force */
1707 fix2 = _mm_add_ps(fix2,tx);
1708 fiy2 = _mm_add_ps(fiy2,ty);
1709 fiz2 = _mm_add_ps(fiz2,tz);
1711 fjx0 = _mm_add_ps(fjx0,tx);
1712 fjy0 = _mm_add_ps(fjy0,ty);
1713 fjz0 = _mm_add_ps(fjz0,tz);
1715 /**************************
1716 * CALCULATE INTERACTIONS *
1717 **************************/
1719 r21 = _mm_mul_ps(rsq21,rinv21);
1721 /* Calculate table index by multiplying r with table scale and truncate to integer */
1722 rt = _mm_mul_ps(r21,vftabscale);
1723 vfitab = _mm_cvttps_epi32(rt);
1724 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1725 vfitab = _mm_slli_epi32(vfitab,2);
1727 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1728 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1729 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1730 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1731 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1732 _MM_TRANSPOSE4_PS(Y,F,G,H);
1733 Heps = _mm_mul_ps(vfeps,H);
1734 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1735 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1736 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
1740 /* Calculate temporary vectorial force */
1741 tx = _mm_mul_ps(fscal,dx21);
1742 ty = _mm_mul_ps(fscal,dy21);
1743 tz = _mm_mul_ps(fscal,dz21);
1745 /* Update vectorial force */
1746 fix2 = _mm_add_ps(fix2,tx);
1747 fiy2 = _mm_add_ps(fiy2,ty);
1748 fiz2 = _mm_add_ps(fiz2,tz);
1750 fjx1 = _mm_add_ps(fjx1,tx);
1751 fjy1 = _mm_add_ps(fjy1,ty);
1752 fjz1 = _mm_add_ps(fjz1,tz);
1754 /**************************
1755 * CALCULATE INTERACTIONS *
1756 **************************/
1758 r22 = _mm_mul_ps(rsq22,rinv22);
1760 /* Calculate table index by multiplying r with table scale and truncate to integer */
1761 rt = _mm_mul_ps(r22,vftabscale);
1762 vfitab = _mm_cvttps_epi32(rt);
1763 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1764 vfitab = _mm_slli_epi32(vfitab,2);
1766 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1767 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1768 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1769 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1770 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1771 _MM_TRANSPOSE4_PS(Y,F,G,H);
1772 Heps = _mm_mul_ps(vfeps,H);
1773 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1774 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1775 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
1779 /* Calculate temporary vectorial force */
1780 tx = _mm_mul_ps(fscal,dx22);
1781 ty = _mm_mul_ps(fscal,dy22);
1782 tz = _mm_mul_ps(fscal,dz22);
1784 /* Update vectorial force */
1785 fix2 = _mm_add_ps(fix2,tx);
1786 fiy2 = _mm_add_ps(fiy2,ty);
1787 fiz2 = _mm_add_ps(fiz2,tz);
1789 fjx2 = _mm_add_ps(fjx2,tx);
1790 fjy2 = _mm_add_ps(fjy2,ty);
1791 fjz2 = _mm_add_ps(fjz2,tz);
1793 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1794 f+j_coord_offsetC,f+j_coord_offsetD,
1795 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1797 /* Inner loop uses 351 flops */
1800 if(jidx<j_index_end)
1803 /* Get j neighbor index, and coordinate index */
1805 jnrB = jjnr[jidx+1];
1806 jnrC = jjnr[jidx+2];
1807 jnrD = jjnr[jidx+3];
1809 /* Sign of each element will be negative for non-real atoms.
1810 * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1811 * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1813 dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1814 jnrA = (jnrA>=0) ? jnrA : 0;
1815 jnrB = (jnrB>=0) ? jnrB : 0;
1816 jnrC = (jnrC>=0) ? jnrC : 0;
1817 jnrD = (jnrD>=0) ? jnrD : 0;
1819 j_coord_offsetA = DIM*jnrA;
1820 j_coord_offsetB = DIM*jnrB;
1821 j_coord_offsetC = DIM*jnrC;
1822 j_coord_offsetD = DIM*jnrD;
1824 /* load j atom coordinates */
1825 gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1826 x+j_coord_offsetC,x+j_coord_offsetD,
1827 &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1829 /* Calculate displacement vector */
1830 dx00 = _mm_sub_ps(ix0,jx0);
1831 dy00 = _mm_sub_ps(iy0,jy0);
1832 dz00 = _mm_sub_ps(iz0,jz0);
1833 dx01 = _mm_sub_ps(ix0,jx1);
1834 dy01 = _mm_sub_ps(iy0,jy1);
1835 dz01 = _mm_sub_ps(iz0,jz1);
1836 dx02 = _mm_sub_ps(ix0,jx2);
1837 dy02 = _mm_sub_ps(iy0,jy2);
1838 dz02 = _mm_sub_ps(iz0,jz2);
1839 dx10 = _mm_sub_ps(ix1,jx0);
1840 dy10 = _mm_sub_ps(iy1,jy0);
1841 dz10 = _mm_sub_ps(iz1,jz0);
1842 dx11 = _mm_sub_ps(ix1,jx1);
1843 dy11 = _mm_sub_ps(iy1,jy1);
1844 dz11 = _mm_sub_ps(iz1,jz1);
1845 dx12 = _mm_sub_ps(ix1,jx2);
1846 dy12 = _mm_sub_ps(iy1,jy2);
1847 dz12 = _mm_sub_ps(iz1,jz2);
1848 dx20 = _mm_sub_ps(ix2,jx0);
1849 dy20 = _mm_sub_ps(iy2,jy0);
1850 dz20 = _mm_sub_ps(iz2,jz0);
1851 dx21 = _mm_sub_ps(ix2,jx1);
1852 dy21 = _mm_sub_ps(iy2,jy1);
1853 dz21 = _mm_sub_ps(iz2,jz1);
1854 dx22 = _mm_sub_ps(ix2,jx2);
1855 dy22 = _mm_sub_ps(iy2,jy2);
1856 dz22 = _mm_sub_ps(iz2,jz2);
1858 /* Calculate squared distance and things based on it */
1859 rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1860 rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1861 rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1862 rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1863 rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1864 rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1865 rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1866 rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1867 rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1869 rinv00 = gmx_mm_invsqrt_ps(rsq00);
1870 rinv01 = gmx_mm_invsqrt_ps(rsq01);
1871 rinv02 = gmx_mm_invsqrt_ps(rsq02);
1872 rinv10 = gmx_mm_invsqrt_ps(rsq10);
1873 rinv11 = gmx_mm_invsqrt_ps(rsq11);
1874 rinv12 = gmx_mm_invsqrt_ps(rsq12);
1875 rinv20 = gmx_mm_invsqrt_ps(rsq20);
1876 rinv21 = gmx_mm_invsqrt_ps(rsq21);
1877 rinv22 = gmx_mm_invsqrt_ps(rsq22);
1879 fjx0 = _mm_setzero_ps();
1880 fjy0 = _mm_setzero_ps();
1881 fjz0 = _mm_setzero_ps();
1882 fjx1 = _mm_setzero_ps();
1883 fjy1 = _mm_setzero_ps();
1884 fjz1 = _mm_setzero_ps();
1885 fjx2 = _mm_setzero_ps();
1886 fjy2 = _mm_setzero_ps();
1887 fjz2 = _mm_setzero_ps();
1889 /**************************
1890 * CALCULATE INTERACTIONS *
1891 **************************/
1893 r00 = _mm_mul_ps(rsq00,rinv00);
1894 r00 = _mm_andnot_ps(dummy_mask,r00);
1896 /* Calculate table index by multiplying r with table scale and truncate to integer */
1897 rt = _mm_mul_ps(r00,vftabscale);
1898 vfitab = _mm_cvttps_epi32(rt);
1899 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1900 vfitab = _mm_slli_epi32(vfitab,2);
1902 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1903 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1904 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1905 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1906 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1907 _MM_TRANSPOSE4_PS(Y,F,G,H);
1908 Heps = _mm_mul_ps(vfeps,H);
1909 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1910 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1911 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1915 fscal = _mm_andnot_ps(dummy_mask,fscal);
1917 /* Calculate temporary vectorial force */
1918 tx = _mm_mul_ps(fscal,dx00);
1919 ty = _mm_mul_ps(fscal,dy00);
1920 tz = _mm_mul_ps(fscal,dz00);
1922 /* Update vectorial force */
1923 fix0 = _mm_add_ps(fix0,tx);
1924 fiy0 = _mm_add_ps(fiy0,ty);
1925 fiz0 = _mm_add_ps(fiz0,tz);
1927 fjx0 = _mm_add_ps(fjx0,tx);
1928 fjy0 = _mm_add_ps(fjy0,ty);
1929 fjz0 = _mm_add_ps(fjz0,tz);
1931 /**************************
1932 * CALCULATE INTERACTIONS *
1933 **************************/
1935 r01 = _mm_mul_ps(rsq01,rinv01);
1936 r01 = _mm_andnot_ps(dummy_mask,r01);
1938 /* Calculate table index by multiplying r with table scale and truncate to integer */
1939 rt = _mm_mul_ps(r01,vftabscale);
1940 vfitab = _mm_cvttps_epi32(rt);
1941 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1942 vfitab = _mm_slli_epi32(vfitab,2);
1944 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1945 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1946 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1947 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1948 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1949 _MM_TRANSPOSE4_PS(Y,F,G,H);
1950 Heps = _mm_mul_ps(vfeps,H);
1951 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1952 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1953 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq01,FF),_mm_mul_ps(vftabscale,rinv01)));
1957 fscal = _mm_andnot_ps(dummy_mask,fscal);
1959 /* Calculate temporary vectorial force */
1960 tx = _mm_mul_ps(fscal,dx01);
1961 ty = _mm_mul_ps(fscal,dy01);
1962 tz = _mm_mul_ps(fscal,dz01);
1964 /* Update vectorial force */
1965 fix0 = _mm_add_ps(fix0,tx);
1966 fiy0 = _mm_add_ps(fiy0,ty);
1967 fiz0 = _mm_add_ps(fiz0,tz);
1969 fjx1 = _mm_add_ps(fjx1,tx);
1970 fjy1 = _mm_add_ps(fjy1,ty);
1971 fjz1 = _mm_add_ps(fjz1,tz);
1973 /**************************
1974 * CALCULATE INTERACTIONS *
1975 **************************/
1977 r02 = _mm_mul_ps(rsq02,rinv02);
1978 r02 = _mm_andnot_ps(dummy_mask,r02);
1980 /* Calculate table index by multiplying r with table scale and truncate to integer */
1981 rt = _mm_mul_ps(r02,vftabscale);
1982 vfitab = _mm_cvttps_epi32(rt);
1983 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1984 vfitab = _mm_slli_epi32(vfitab,2);
1986 /* CUBIC SPLINE TABLE ELECTROSTATICS */
1987 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1988 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1989 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1990 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1991 _MM_TRANSPOSE4_PS(Y,F,G,H);
1992 Heps = _mm_mul_ps(vfeps,H);
1993 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1994 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1995 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq02,FF),_mm_mul_ps(vftabscale,rinv02)));
1999 fscal = _mm_andnot_ps(dummy_mask,fscal);
2001 /* Calculate temporary vectorial force */
2002 tx = _mm_mul_ps(fscal,dx02);
2003 ty = _mm_mul_ps(fscal,dy02);
2004 tz = _mm_mul_ps(fscal,dz02);
2006 /* Update vectorial force */
2007 fix0 = _mm_add_ps(fix0,tx);
2008 fiy0 = _mm_add_ps(fiy0,ty);
2009 fiz0 = _mm_add_ps(fiz0,tz);
2011 fjx2 = _mm_add_ps(fjx2,tx);
2012 fjy2 = _mm_add_ps(fjy2,ty);
2013 fjz2 = _mm_add_ps(fjz2,tz);
2015 /**************************
2016 * CALCULATE INTERACTIONS *
2017 **************************/
2019 r10 = _mm_mul_ps(rsq10,rinv10);
2020 r10 = _mm_andnot_ps(dummy_mask,r10);
2022 /* Calculate table index by multiplying r with table scale and truncate to integer */
2023 rt = _mm_mul_ps(r10,vftabscale);
2024 vfitab = _mm_cvttps_epi32(rt);
2025 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2026 vfitab = _mm_slli_epi32(vfitab,2);
2028 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2029 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2030 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2031 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2032 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2033 _MM_TRANSPOSE4_PS(Y,F,G,H);
2034 Heps = _mm_mul_ps(vfeps,H);
2035 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2036 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2037 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
2041 fscal = _mm_andnot_ps(dummy_mask,fscal);
2043 /* Calculate temporary vectorial force */
2044 tx = _mm_mul_ps(fscal,dx10);
2045 ty = _mm_mul_ps(fscal,dy10);
2046 tz = _mm_mul_ps(fscal,dz10);
2048 /* Update vectorial force */
2049 fix1 = _mm_add_ps(fix1,tx);
2050 fiy1 = _mm_add_ps(fiy1,ty);
2051 fiz1 = _mm_add_ps(fiz1,tz);
2053 fjx0 = _mm_add_ps(fjx0,tx);
2054 fjy0 = _mm_add_ps(fjy0,ty);
2055 fjz0 = _mm_add_ps(fjz0,tz);
2057 /**************************
2058 * CALCULATE INTERACTIONS *
2059 **************************/
2061 r11 = _mm_mul_ps(rsq11,rinv11);
2062 r11 = _mm_andnot_ps(dummy_mask,r11);
2064 /* Calculate table index by multiplying r with table scale and truncate to integer */
2065 rt = _mm_mul_ps(r11,vftabscale);
2066 vfitab = _mm_cvttps_epi32(rt);
2067 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2068 vfitab = _mm_slli_epi32(vfitab,2);
2070 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2071 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2072 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2073 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2074 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2075 _MM_TRANSPOSE4_PS(Y,F,G,H);
2076 Heps = _mm_mul_ps(vfeps,H);
2077 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2078 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2079 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq11,FF),_mm_mul_ps(vftabscale,rinv11)));
2083 fscal = _mm_andnot_ps(dummy_mask,fscal);
2085 /* Calculate temporary vectorial force */
2086 tx = _mm_mul_ps(fscal,dx11);
2087 ty = _mm_mul_ps(fscal,dy11);
2088 tz = _mm_mul_ps(fscal,dz11);
2090 /* Update vectorial force */
2091 fix1 = _mm_add_ps(fix1,tx);
2092 fiy1 = _mm_add_ps(fiy1,ty);
2093 fiz1 = _mm_add_ps(fiz1,tz);
2095 fjx1 = _mm_add_ps(fjx1,tx);
2096 fjy1 = _mm_add_ps(fjy1,ty);
2097 fjz1 = _mm_add_ps(fjz1,tz);
2099 /**************************
2100 * CALCULATE INTERACTIONS *
2101 **************************/
2103 r12 = _mm_mul_ps(rsq12,rinv12);
2104 r12 = _mm_andnot_ps(dummy_mask,r12);
2106 /* Calculate table index by multiplying r with table scale and truncate to integer */
2107 rt = _mm_mul_ps(r12,vftabscale);
2108 vfitab = _mm_cvttps_epi32(rt);
2109 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2110 vfitab = _mm_slli_epi32(vfitab,2);
2112 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2113 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2114 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2115 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2116 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2117 _MM_TRANSPOSE4_PS(Y,F,G,H);
2118 Heps = _mm_mul_ps(vfeps,H);
2119 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2120 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2121 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq12,FF),_mm_mul_ps(vftabscale,rinv12)));
2125 fscal = _mm_andnot_ps(dummy_mask,fscal);
2127 /* Calculate temporary vectorial force */
2128 tx = _mm_mul_ps(fscal,dx12);
2129 ty = _mm_mul_ps(fscal,dy12);
2130 tz = _mm_mul_ps(fscal,dz12);
2132 /* Update vectorial force */
2133 fix1 = _mm_add_ps(fix1,tx);
2134 fiy1 = _mm_add_ps(fiy1,ty);
2135 fiz1 = _mm_add_ps(fiz1,tz);
2137 fjx2 = _mm_add_ps(fjx2,tx);
2138 fjy2 = _mm_add_ps(fjy2,ty);
2139 fjz2 = _mm_add_ps(fjz2,tz);
2141 /**************************
2142 * CALCULATE INTERACTIONS *
2143 **************************/
2145 r20 = _mm_mul_ps(rsq20,rinv20);
2146 r20 = _mm_andnot_ps(dummy_mask,r20);
2148 /* Calculate table index by multiplying r with table scale and truncate to integer */
2149 rt = _mm_mul_ps(r20,vftabscale);
2150 vfitab = _mm_cvttps_epi32(rt);
2151 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2152 vfitab = _mm_slli_epi32(vfitab,2);
2154 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2155 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2156 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2157 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2158 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2159 _MM_TRANSPOSE4_PS(Y,F,G,H);
2160 Heps = _mm_mul_ps(vfeps,H);
2161 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2162 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2163 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
2167 fscal = _mm_andnot_ps(dummy_mask,fscal);
2169 /* Calculate temporary vectorial force */
2170 tx = _mm_mul_ps(fscal,dx20);
2171 ty = _mm_mul_ps(fscal,dy20);
2172 tz = _mm_mul_ps(fscal,dz20);
2174 /* Update vectorial force */
2175 fix2 = _mm_add_ps(fix2,tx);
2176 fiy2 = _mm_add_ps(fiy2,ty);
2177 fiz2 = _mm_add_ps(fiz2,tz);
2179 fjx0 = _mm_add_ps(fjx0,tx);
2180 fjy0 = _mm_add_ps(fjy0,ty);
2181 fjz0 = _mm_add_ps(fjz0,tz);
2183 /**************************
2184 * CALCULATE INTERACTIONS *
2185 **************************/
2187 r21 = _mm_mul_ps(rsq21,rinv21);
2188 r21 = _mm_andnot_ps(dummy_mask,r21);
2190 /* Calculate table index by multiplying r with table scale and truncate to integer */
2191 rt = _mm_mul_ps(r21,vftabscale);
2192 vfitab = _mm_cvttps_epi32(rt);
2193 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2194 vfitab = _mm_slli_epi32(vfitab,2);
2196 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2197 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2198 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2199 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2200 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2201 _MM_TRANSPOSE4_PS(Y,F,G,H);
2202 Heps = _mm_mul_ps(vfeps,H);
2203 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2204 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2205 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq21,FF),_mm_mul_ps(vftabscale,rinv21)));
2209 fscal = _mm_andnot_ps(dummy_mask,fscal);
2211 /* Calculate temporary vectorial force */
2212 tx = _mm_mul_ps(fscal,dx21);
2213 ty = _mm_mul_ps(fscal,dy21);
2214 tz = _mm_mul_ps(fscal,dz21);
2216 /* Update vectorial force */
2217 fix2 = _mm_add_ps(fix2,tx);
2218 fiy2 = _mm_add_ps(fiy2,ty);
2219 fiz2 = _mm_add_ps(fiz2,tz);
2221 fjx1 = _mm_add_ps(fjx1,tx);
2222 fjy1 = _mm_add_ps(fjy1,ty);
2223 fjz1 = _mm_add_ps(fjz1,tz);
2225 /**************************
2226 * CALCULATE INTERACTIONS *
2227 **************************/
2229 r22 = _mm_mul_ps(rsq22,rinv22);
2230 r22 = _mm_andnot_ps(dummy_mask,r22);
2232 /* Calculate table index by multiplying r with table scale and truncate to integer */
2233 rt = _mm_mul_ps(r22,vftabscale);
2234 vfitab = _mm_cvttps_epi32(rt);
2235 vfeps = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
2236 vfitab = _mm_slli_epi32(vfitab,2);
2238 /* CUBIC SPLINE TABLE ELECTROSTATICS */
2239 Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
2240 F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
2241 G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
2242 H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
2243 _MM_TRANSPOSE4_PS(Y,F,G,H);
2244 Heps = _mm_mul_ps(vfeps,H);
2245 Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
2246 FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
2247 felec = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq22,FF),_mm_mul_ps(vftabscale,rinv22)));
2251 fscal = _mm_andnot_ps(dummy_mask,fscal);
2253 /* Calculate temporary vectorial force */
2254 tx = _mm_mul_ps(fscal,dx22);
2255 ty = _mm_mul_ps(fscal,dy22);
2256 tz = _mm_mul_ps(fscal,dz22);
2258 /* Update vectorial force */
2259 fix2 = _mm_add_ps(fix2,tx);
2260 fiy2 = _mm_add_ps(fiy2,ty);
2261 fiz2 = _mm_add_ps(fiz2,tz);
2263 fjx2 = _mm_add_ps(fjx2,tx);
2264 fjy2 = _mm_add_ps(fjy2,ty);
2265 fjz2 = _mm_add_ps(fjz2,tz);
2267 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
2268 f+j_coord_offsetC,f+j_coord_offsetD,
2269 fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2271 /* Inner loop uses 360 flops */
2274 /* End of innermost loop */
2276 gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2277 f+i_coord_offset,fshift+i_shift_offset);
2279 /* Increment number of inner iterations */
2280 inneriter += j_index_end - j_index_start;
2282 /* Outer loop uses 27 flops */
2285 /* Increment number of outer iterations */
2288 /* Update outer/inner flops */
2290 inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3W3_F,outeriter*27 + inneriter*360);