File: | gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRF_VdwCSTab_GeomW3W3_sse4_1_single.c |
Location: | line 1122, column 5 |
Description: | Value stored to 'crf' is never read |
1 | /* |
2 | * This file is part of the GROMACS molecular simulation package. |
3 | * |
4 | * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by |
5 | * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl, |
6 | * and including many others, as listed in the AUTHORS file in the |
7 | * top-level source directory and at http://www.gromacs.org. |
8 | * |
9 | * GROMACS is free software; you can redistribute it and/or |
10 | * modify it under the terms of the GNU Lesser General Public License |
11 | * as published by the Free Software Foundation; either version 2.1 |
12 | * of the License, or (at your option) any later version. |
13 | * |
14 | * GROMACS is distributed in the hope that it will be useful, |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
17 | * Lesser General Public License for more details. |
18 | * |
19 | * You should have received a copy of the GNU Lesser General Public |
20 | * License along with GROMACS; if not, see |
21 | * http://www.gnu.org/licenses, or write to the Free Software Foundation, |
22 | * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. |
23 | * |
24 | * If you want to redistribute modifications to GROMACS, please |
25 | * consider that scientific software is very special. Version |
26 | * control is crucial - bugs must be traceable. We will be happy to |
27 | * consider code for inclusion in the official distribution, but |
28 | * derived work must not be called official GROMACS. Details are found |
29 | * in the README & COPYING files - if they are missing, get the |
30 | * official version at http://www.gromacs.org. |
31 | * |
32 | * To help us fund GROMACS development, we humbly ask that you cite |
33 | * the research papers on the package. Check out http://www.gromacs.org. |
34 | */ |
35 | /* |
36 | * Note: this file was generated by the GROMACS sse4_1_single kernel generator. |
37 | */ |
38 | #ifdef HAVE_CONFIG_H1 |
39 | #include <config.h> |
40 | #endif |
41 | |
42 | #include <math.h> |
43 | |
44 | #include "../nb_kernel.h" |
45 | #include "types/simple.h" |
46 | #include "gromacs/math/vec.h" |
47 | #include "nrnb.h" |
48 | |
49 | #include "gromacs/simd/math_x86_sse4_1_single.h" |
50 | #include "kernelutil_x86_sse4_1_single.h" |
51 | |
52 | /* |
53 | * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW3W3_VF_sse4_1_single |
54 | * Electrostatics interaction: ReactionField |
55 | * VdW interaction: CubicSplineTable |
56 | * Geometry: Water3-Water3 |
57 | * Calculate force/pot: PotentialAndForce |
58 | */ |
59 | void |
60 | nb_kernel_ElecRF_VdwCSTab_GeomW3W3_VF_sse4_1_single |
61 | (t_nblist * gmx_restrict nlist, |
62 | rvec * gmx_restrict xx, |
63 | rvec * gmx_restrict ff, |
64 | t_forcerec * gmx_restrict fr, |
65 | t_mdatoms * gmx_restrict mdatoms, |
66 | nb_kernel_data_t gmx_unused__attribute__ ((unused)) * gmx_restrict kernel_data, |
67 | t_nrnb * gmx_restrict nrnb) |
68 | { |
69 | /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or |
70 | * just 0 for non-waters. |
71 | * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different |
72 | * jnr indices corresponding to data put in the four positions in the SIMD register. |
73 | */ |
74 | int i_shift_offset,i_coord_offset,outeriter,inneriter; |
75 | int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx; |
76 | int jnrA,jnrB,jnrC,jnrD; |
77 | int jnrlistA,jnrlistB,jnrlistC,jnrlistD; |
78 | int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD; |
79 | int *iinr,*jindex,*jjnr,*shiftidx,*gid; |
80 | real rcutoff_scalar; |
81 | real *shiftvec,*fshift,*x,*f; |
82 | real *fjptrA,*fjptrB,*fjptrC,*fjptrD; |
83 | real scratch[4*DIM3]; |
84 | __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall; |
85 | int vdwioffset0; |
86 | __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0; |
87 | int vdwioffset1; |
88 | __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1; |
89 | int vdwioffset2; |
90 | __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2; |
91 | int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D; |
92 | __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0; |
93 | int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D; |
94 | __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1; |
95 | int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D; |
96 | __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2; |
97 | __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00; |
98 | __m128 dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01; |
99 | __m128 dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02; |
100 | __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10; |
101 | __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11; |
102 | __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12; |
103 | __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20; |
104 | __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21; |
105 | __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22; |
106 | __m128 velec,felec,velecsum,facel,crf,krf,krf2; |
107 | real *charge; |
108 | int nvdwtype; |
109 | __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6; |
110 | int *vdwtype; |
111 | real *vdwparam; |
112 | __m128 one_sixth = _mm_set1_ps(1.0/6.0); |
113 | __m128 one_twelfth = _mm_set1_ps(1.0/12.0); |
114 | __m128i vfitab; |
115 | __m128i ifour = _mm_set1_epi32(4); |
116 | __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF; |
117 | real *vftab; |
118 | __m128 dummy_mask,cutoff_mask; |
119 | __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) ); |
120 | __m128 one = _mm_set1_ps(1.0); |
121 | __m128 two = _mm_set1_ps(2.0); |
122 | x = xx[0]; |
123 | f = ff[0]; |
124 | |
125 | nri = nlist->nri; |
126 | iinr = nlist->iinr; |
127 | jindex = nlist->jindex; |
128 | jjnr = nlist->jjnr; |
129 | shiftidx = nlist->shift; |
130 | gid = nlist->gid; |
131 | shiftvec = fr->shift_vec[0]; |
132 | fshift = fr->fshift[0]; |
133 | facel = _mm_set1_ps(fr->epsfac); |
134 | charge = mdatoms->chargeA; |
135 | krf = _mm_set1_ps(fr->ic->k_rf); |
136 | krf2 = _mm_set1_ps(fr->ic->k_rf*2.0); |
137 | crf = _mm_set1_ps(fr->ic->c_rf); |
138 | nvdwtype = fr->ntype; |
139 | vdwparam = fr->nbfp; |
140 | vdwtype = mdatoms->typeA; |
141 | |
142 | vftab = kernel_data->table_vdw->data; |
143 | vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale); |
144 | |
145 | /* Setup water-specific parameters */ |
146 | inr = nlist->iinr[0]; |
147 | iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0])); |
148 | iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1])); |
149 | iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2])); |
150 | vdwioffset0 = 2*nvdwtype*vdwtype[inr+0]; |
151 | |
152 | jq0 = _mm_set1_ps(charge[inr+0]); |
153 | jq1 = _mm_set1_ps(charge[inr+1]); |
154 | jq2 = _mm_set1_ps(charge[inr+2]); |
155 | vdwjidx0A = 2*vdwtype[inr+0]; |
156 | qq00 = _mm_mul_ps(iq0,jq0); |
157 | c6_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]); |
158 | c12_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]); |
159 | qq01 = _mm_mul_ps(iq0,jq1); |
160 | qq02 = _mm_mul_ps(iq0,jq2); |
161 | qq10 = _mm_mul_ps(iq1,jq0); |
162 | qq11 = _mm_mul_ps(iq1,jq1); |
163 | qq12 = _mm_mul_ps(iq1,jq2); |
164 | qq20 = _mm_mul_ps(iq2,jq0); |
165 | qq21 = _mm_mul_ps(iq2,jq1); |
166 | qq22 = _mm_mul_ps(iq2,jq2); |
167 | |
168 | /* Avoid stupid compiler warnings */ |
169 | jnrA = jnrB = jnrC = jnrD = 0; |
170 | j_coord_offsetA = 0; |
171 | j_coord_offsetB = 0; |
172 | j_coord_offsetC = 0; |
173 | j_coord_offsetD = 0; |
174 | |
175 | outeriter = 0; |
176 | inneriter = 0; |
177 | |
178 | for(iidx=0;iidx<4*DIM3;iidx++) |
179 | { |
180 | scratch[iidx] = 0.0; |
181 | } |
182 | |
183 | /* Start outer loop over neighborlists */ |
184 | for(iidx=0; iidx<nri; iidx++) |
185 | { |
186 | /* Load shift vector for this list */ |
187 | i_shift_offset = DIM3*shiftidx[iidx]; |
188 | |
189 | /* Load limits for loop over neighbors */ |
190 | j_index_start = jindex[iidx]; |
191 | j_index_end = jindex[iidx+1]; |
192 | |
193 | /* Get outer coordinate index */ |
194 | inr = iinr[iidx]; |
195 | i_coord_offset = DIM3*inr; |
196 | |
197 | /* Load i particle coords and add shift vector */ |
198 | gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset, |
199 | &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2); |
200 | |
201 | fix0 = _mm_setzero_ps(); |
202 | fiy0 = _mm_setzero_ps(); |
203 | fiz0 = _mm_setzero_ps(); |
204 | fix1 = _mm_setzero_ps(); |
205 | fiy1 = _mm_setzero_ps(); |
206 | fiz1 = _mm_setzero_ps(); |
207 | fix2 = _mm_setzero_ps(); |
208 | fiy2 = _mm_setzero_ps(); |
209 | fiz2 = _mm_setzero_ps(); |
210 | |
211 | /* Reset potential sums */ |
212 | velecsum = _mm_setzero_ps(); |
213 | vvdwsum = _mm_setzero_ps(); |
214 | |
215 | /* Start inner kernel loop */ |
216 | for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4) |
217 | { |
218 | |
219 | /* Get j neighbor index, and coordinate index */ |
220 | jnrA = jjnr[jidx]; |
221 | jnrB = jjnr[jidx+1]; |
222 | jnrC = jjnr[jidx+2]; |
223 | jnrD = jjnr[jidx+3]; |
224 | j_coord_offsetA = DIM3*jnrA; |
225 | j_coord_offsetB = DIM3*jnrB; |
226 | j_coord_offsetC = DIM3*jnrC; |
227 | j_coord_offsetD = DIM3*jnrD; |
228 | |
229 | /* load j atom coordinates */ |
230 | gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
231 | x+j_coord_offsetC,x+j_coord_offsetD, |
232 | &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2); |
233 | |
234 | /* Calculate displacement vector */ |
235 | dx00 = _mm_sub_ps(ix0,jx0); |
236 | dy00 = _mm_sub_ps(iy0,jy0); |
237 | dz00 = _mm_sub_ps(iz0,jz0); |
238 | dx01 = _mm_sub_ps(ix0,jx1); |
239 | dy01 = _mm_sub_ps(iy0,jy1); |
240 | dz01 = _mm_sub_ps(iz0,jz1); |
241 | dx02 = _mm_sub_ps(ix0,jx2); |
242 | dy02 = _mm_sub_ps(iy0,jy2); |
243 | dz02 = _mm_sub_ps(iz0,jz2); |
244 | dx10 = _mm_sub_ps(ix1,jx0); |
245 | dy10 = _mm_sub_ps(iy1,jy0); |
246 | dz10 = _mm_sub_ps(iz1,jz0); |
247 | dx11 = _mm_sub_ps(ix1,jx1); |
248 | dy11 = _mm_sub_ps(iy1,jy1); |
249 | dz11 = _mm_sub_ps(iz1,jz1); |
250 | dx12 = _mm_sub_ps(ix1,jx2); |
251 | dy12 = _mm_sub_ps(iy1,jy2); |
252 | dz12 = _mm_sub_ps(iz1,jz2); |
253 | dx20 = _mm_sub_ps(ix2,jx0); |
254 | dy20 = _mm_sub_ps(iy2,jy0); |
255 | dz20 = _mm_sub_ps(iz2,jz0); |
256 | dx21 = _mm_sub_ps(ix2,jx1); |
257 | dy21 = _mm_sub_ps(iy2,jy1); |
258 | dz21 = _mm_sub_ps(iz2,jz1); |
259 | dx22 = _mm_sub_ps(ix2,jx2); |
260 | dy22 = _mm_sub_ps(iy2,jy2); |
261 | dz22 = _mm_sub_ps(iz2,jz2); |
262 | |
263 | /* Calculate squared distance and things based on it */ |
264 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
265 | rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01); |
266 | rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02); |
267 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
268 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
269 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
270 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
271 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
272 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
273 | |
274 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
275 | rinv01 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq01); |
276 | rinv02 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq02); |
277 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
278 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
279 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
280 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
281 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
282 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
283 | |
284 | rinvsq00 = _mm_mul_ps(rinv00,rinv00); |
285 | rinvsq01 = _mm_mul_ps(rinv01,rinv01); |
286 | rinvsq02 = _mm_mul_ps(rinv02,rinv02); |
287 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
288 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
289 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
290 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
291 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
292 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
293 | |
294 | fjx0 = _mm_setzero_ps(); |
295 | fjy0 = _mm_setzero_ps(); |
296 | fjz0 = _mm_setzero_ps(); |
297 | fjx1 = _mm_setzero_ps(); |
298 | fjy1 = _mm_setzero_ps(); |
299 | fjz1 = _mm_setzero_ps(); |
300 | fjx2 = _mm_setzero_ps(); |
301 | fjy2 = _mm_setzero_ps(); |
302 | fjz2 = _mm_setzero_ps(); |
303 | |
304 | /************************** |
305 | * CALCULATE INTERACTIONS * |
306 | **************************/ |
307 | |
308 | r00 = _mm_mul_ps(rsq00,rinv00); |
309 | |
310 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
311 | rt = _mm_mul_ps(r00,vftabscale); |
312 | vfitab = _mm_cvttps_epi32(rt); |
313 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
314 | vfitab = _mm_slli_epi32(vfitab,3); |
315 | |
316 | /* REACTION-FIELD ELECTROSTATICS */ |
317 | velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf)); |
318 | felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2)); |
319 | |
320 | /* CUBIC SPLINE TABLE DISPERSION */ |
321 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
322 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
323 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
324 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
325 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
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 | vvdw6 = _mm_mul_ps(c6_00,VV); |
330 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
331 | fvdw6 = _mm_mul_ps(c6_00,FF); |
332 | |
333 | /* CUBIC SPLINE TABLE REPULSION */ |
334 | vfitab = _mm_add_epi32(vfitab,ifour); |
335 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
336 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
337 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
338 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
339 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
340 | Heps = _mm_mul_ps(vfeps,H); |
341 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
342 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
343 | vvdw12 = _mm_mul_ps(c12_00,VV); |
344 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
345 | fvdw12 = _mm_mul_ps(c12_00,FF); |
346 | vvdw = _mm_add_ps(vvdw12,vvdw6); |
347 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
348 | |
349 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
350 | velecsum = _mm_add_ps(velecsum,velec); |
351 | vvdwsum = _mm_add_ps(vvdwsum,vvdw); |
352 | |
353 | fscal = _mm_add_ps(felec,fvdw); |
354 | |
355 | /* Calculate temporary vectorial force */ |
356 | tx = _mm_mul_ps(fscal,dx00); |
357 | ty = _mm_mul_ps(fscal,dy00); |
358 | tz = _mm_mul_ps(fscal,dz00); |
359 | |
360 | /* Update vectorial force */ |
361 | fix0 = _mm_add_ps(fix0,tx); |
362 | fiy0 = _mm_add_ps(fiy0,ty); |
363 | fiz0 = _mm_add_ps(fiz0,tz); |
364 | |
365 | fjx0 = _mm_add_ps(fjx0,tx); |
366 | fjy0 = _mm_add_ps(fjy0,ty); |
367 | fjz0 = _mm_add_ps(fjz0,tz); |
368 | |
369 | /************************** |
370 | * CALCULATE INTERACTIONS * |
371 | **************************/ |
372 | |
373 | /* REACTION-FIELD ELECTROSTATICS */ |
374 | velec = _mm_mul_ps(qq01,_mm_sub_ps(_mm_add_ps(rinv01,_mm_mul_ps(krf,rsq01)),crf)); |
375 | felec = _mm_mul_ps(qq01,_mm_sub_ps(_mm_mul_ps(rinv01,rinvsq01),krf2)); |
376 | |
377 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
378 | velecsum = _mm_add_ps(velecsum,velec); |
379 | |
380 | fscal = felec; |
381 | |
382 | /* Calculate temporary vectorial force */ |
383 | tx = _mm_mul_ps(fscal,dx01); |
384 | ty = _mm_mul_ps(fscal,dy01); |
385 | tz = _mm_mul_ps(fscal,dz01); |
386 | |
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); |
391 | |
392 | fjx1 = _mm_add_ps(fjx1,tx); |
393 | fjy1 = _mm_add_ps(fjy1,ty); |
394 | fjz1 = _mm_add_ps(fjz1,tz); |
395 | |
396 | /************************** |
397 | * CALCULATE INTERACTIONS * |
398 | **************************/ |
399 | |
400 | /* REACTION-FIELD ELECTROSTATICS */ |
401 | velec = _mm_mul_ps(qq02,_mm_sub_ps(_mm_add_ps(rinv02,_mm_mul_ps(krf,rsq02)),crf)); |
402 | felec = _mm_mul_ps(qq02,_mm_sub_ps(_mm_mul_ps(rinv02,rinvsq02),krf2)); |
403 | |
404 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
405 | velecsum = _mm_add_ps(velecsum,velec); |
406 | |
407 | fscal = felec; |
408 | |
409 | /* Calculate temporary vectorial force */ |
410 | tx = _mm_mul_ps(fscal,dx02); |
411 | ty = _mm_mul_ps(fscal,dy02); |
412 | tz = _mm_mul_ps(fscal,dz02); |
413 | |
414 | /* Update vectorial force */ |
415 | fix0 = _mm_add_ps(fix0,tx); |
416 | fiy0 = _mm_add_ps(fiy0,ty); |
417 | fiz0 = _mm_add_ps(fiz0,tz); |
418 | |
419 | fjx2 = _mm_add_ps(fjx2,tx); |
420 | fjy2 = _mm_add_ps(fjy2,ty); |
421 | fjz2 = _mm_add_ps(fjz2,tz); |
422 | |
423 | /************************** |
424 | * CALCULATE INTERACTIONS * |
425 | **************************/ |
426 | |
427 | /* REACTION-FIELD ELECTROSTATICS */ |
428 | velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf)); |
429 | felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2)); |
430 | |
431 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
432 | velecsum = _mm_add_ps(velecsum,velec); |
433 | |
434 | fscal = felec; |
435 | |
436 | /* Calculate temporary vectorial force */ |
437 | tx = _mm_mul_ps(fscal,dx10); |
438 | ty = _mm_mul_ps(fscal,dy10); |
439 | tz = _mm_mul_ps(fscal,dz10); |
440 | |
441 | /* Update vectorial force */ |
442 | fix1 = _mm_add_ps(fix1,tx); |
443 | fiy1 = _mm_add_ps(fiy1,ty); |
444 | fiz1 = _mm_add_ps(fiz1,tz); |
445 | |
446 | fjx0 = _mm_add_ps(fjx0,tx); |
447 | fjy0 = _mm_add_ps(fjy0,ty); |
448 | fjz0 = _mm_add_ps(fjz0,tz); |
449 | |
450 | /************************** |
451 | * CALCULATE INTERACTIONS * |
452 | **************************/ |
453 | |
454 | /* REACTION-FIELD ELECTROSTATICS */ |
455 | velec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_add_ps(rinv11,_mm_mul_ps(krf,rsq11)),crf)); |
456 | felec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_mul_ps(rinv11,rinvsq11),krf2)); |
457 | |
458 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
459 | velecsum = _mm_add_ps(velecsum,velec); |
460 | |
461 | fscal = felec; |
462 | |
463 | /* Calculate temporary vectorial force */ |
464 | tx = _mm_mul_ps(fscal,dx11); |
465 | ty = _mm_mul_ps(fscal,dy11); |
466 | tz = _mm_mul_ps(fscal,dz11); |
467 | |
468 | /* Update vectorial force */ |
469 | fix1 = _mm_add_ps(fix1,tx); |
470 | fiy1 = _mm_add_ps(fiy1,ty); |
471 | fiz1 = _mm_add_ps(fiz1,tz); |
472 | |
473 | fjx1 = _mm_add_ps(fjx1,tx); |
474 | fjy1 = _mm_add_ps(fjy1,ty); |
475 | fjz1 = _mm_add_ps(fjz1,tz); |
476 | |
477 | /************************** |
478 | * CALCULATE INTERACTIONS * |
479 | **************************/ |
480 | |
481 | /* REACTION-FIELD ELECTROSTATICS */ |
482 | velec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_add_ps(rinv12,_mm_mul_ps(krf,rsq12)),crf)); |
483 | felec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_mul_ps(rinv12,rinvsq12),krf2)); |
484 | |
485 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
486 | velecsum = _mm_add_ps(velecsum,velec); |
487 | |
488 | fscal = felec; |
489 | |
490 | /* Calculate temporary vectorial force */ |
491 | tx = _mm_mul_ps(fscal,dx12); |
492 | ty = _mm_mul_ps(fscal,dy12); |
493 | tz = _mm_mul_ps(fscal,dz12); |
494 | |
495 | /* Update vectorial force */ |
496 | fix1 = _mm_add_ps(fix1,tx); |
497 | fiy1 = _mm_add_ps(fiy1,ty); |
498 | fiz1 = _mm_add_ps(fiz1,tz); |
499 | |
500 | fjx2 = _mm_add_ps(fjx2,tx); |
501 | fjy2 = _mm_add_ps(fjy2,ty); |
502 | fjz2 = _mm_add_ps(fjz2,tz); |
503 | |
504 | /************************** |
505 | * CALCULATE INTERACTIONS * |
506 | **************************/ |
507 | |
508 | /* REACTION-FIELD ELECTROSTATICS */ |
509 | velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf)); |
510 | felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2)); |
511 | |
512 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
513 | velecsum = _mm_add_ps(velecsum,velec); |
514 | |
515 | fscal = felec; |
516 | |
517 | /* Calculate temporary vectorial force */ |
518 | tx = _mm_mul_ps(fscal,dx20); |
519 | ty = _mm_mul_ps(fscal,dy20); |
520 | tz = _mm_mul_ps(fscal,dz20); |
521 | |
522 | /* Update vectorial force */ |
523 | fix2 = _mm_add_ps(fix2,tx); |
524 | fiy2 = _mm_add_ps(fiy2,ty); |
525 | fiz2 = _mm_add_ps(fiz2,tz); |
526 | |
527 | fjx0 = _mm_add_ps(fjx0,tx); |
528 | fjy0 = _mm_add_ps(fjy0,ty); |
529 | fjz0 = _mm_add_ps(fjz0,tz); |
530 | |
531 | /************************** |
532 | * CALCULATE INTERACTIONS * |
533 | **************************/ |
534 | |
535 | /* REACTION-FIELD ELECTROSTATICS */ |
536 | velec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_add_ps(rinv21,_mm_mul_ps(krf,rsq21)),crf)); |
537 | felec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_mul_ps(rinv21,rinvsq21),krf2)); |
538 | |
539 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
540 | velecsum = _mm_add_ps(velecsum,velec); |
541 | |
542 | fscal = felec; |
543 | |
544 | /* Calculate temporary vectorial force */ |
545 | tx = _mm_mul_ps(fscal,dx21); |
546 | ty = _mm_mul_ps(fscal,dy21); |
547 | tz = _mm_mul_ps(fscal,dz21); |
548 | |
549 | /* Update vectorial force */ |
550 | fix2 = _mm_add_ps(fix2,tx); |
551 | fiy2 = _mm_add_ps(fiy2,ty); |
552 | fiz2 = _mm_add_ps(fiz2,tz); |
553 | |
554 | fjx1 = _mm_add_ps(fjx1,tx); |
555 | fjy1 = _mm_add_ps(fjy1,ty); |
556 | fjz1 = _mm_add_ps(fjz1,tz); |
557 | |
558 | /************************** |
559 | * CALCULATE INTERACTIONS * |
560 | **************************/ |
561 | |
562 | /* REACTION-FIELD ELECTROSTATICS */ |
563 | velec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_add_ps(rinv22,_mm_mul_ps(krf,rsq22)),crf)); |
564 | felec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_mul_ps(rinv22,rinvsq22),krf2)); |
565 | |
566 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
567 | velecsum = _mm_add_ps(velecsum,velec); |
568 | |
569 | fscal = felec; |
570 | |
571 | /* Calculate temporary vectorial force */ |
572 | tx = _mm_mul_ps(fscal,dx22); |
573 | ty = _mm_mul_ps(fscal,dy22); |
574 | tz = _mm_mul_ps(fscal,dz22); |
575 | |
576 | /* Update vectorial force */ |
577 | fix2 = _mm_add_ps(fix2,tx); |
578 | fiy2 = _mm_add_ps(fiy2,ty); |
579 | fiz2 = _mm_add_ps(fiz2,tz); |
580 | |
581 | fjx2 = _mm_add_ps(fjx2,tx); |
582 | fjy2 = _mm_add_ps(fjy2,ty); |
583 | fjz2 = _mm_add_ps(fjz2,tz); |
584 | |
585 | fjptrA = f+j_coord_offsetA; |
586 | fjptrB = f+j_coord_offsetB; |
587 | fjptrC = f+j_coord_offsetC; |
588 | fjptrD = f+j_coord_offsetD; |
589 | |
590 | gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD, |
591 | fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2); |
592 | |
593 | /* Inner loop uses 323 flops */ |
594 | } |
595 | |
596 | if(jidx<j_index_end) |
597 | { |
598 | |
599 | /* Get j neighbor index, and coordinate index */ |
600 | jnrlistA = jjnr[jidx]; |
601 | jnrlistB = jjnr[jidx+1]; |
602 | jnrlistC = jjnr[jidx+2]; |
603 | jnrlistD = jjnr[jidx+3]; |
604 | /* Sign of each element will be negative for non-real atoms. |
605 | * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones, |
606 | * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries. |
607 | */ |
608 | dummy_mask = gmx_mm_castsi128_ps_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())); |
609 | jnrA = (jnrlistA>=0) ? jnrlistA : 0; |
610 | jnrB = (jnrlistB>=0) ? jnrlistB : 0; |
611 | jnrC = (jnrlistC>=0) ? jnrlistC : 0; |
612 | jnrD = (jnrlistD>=0) ? jnrlistD : 0; |
613 | j_coord_offsetA = DIM3*jnrA; |
614 | j_coord_offsetB = DIM3*jnrB; |
615 | j_coord_offsetC = DIM3*jnrC; |
616 | j_coord_offsetD = DIM3*jnrD; |
617 | |
618 | /* load j atom coordinates */ |
619 | gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
620 | x+j_coord_offsetC,x+j_coord_offsetD, |
621 | &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2); |
622 | |
623 | /* Calculate displacement vector */ |
624 | dx00 = _mm_sub_ps(ix0,jx0); |
625 | dy00 = _mm_sub_ps(iy0,jy0); |
626 | dz00 = _mm_sub_ps(iz0,jz0); |
627 | dx01 = _mm_sub_ps(ix0,jx1); |
628 | dy01 = _mm_sub_ps(iy0,jy1); |
629 | dz01 = _mm_sub_ps(iz0,jz1); |
630 | dx02 = _mm_sub_ps(ix0,jx2); |
631 | dy02 = _mm_sub_ps(iy0,jy2); |
632 | dz02 = _mm_sub_ps(iz0,jz2); |
633 | dx10 = _mm_sub_ps(ix1,jx0); |
634 | dy10 = _mm_sub_ps(iy1,jy0); |
635 | dz10 = _mm_sub_ps(iz1,jz0); |
636 | dx11 = _mm_sub_ps(ix1,jx1); |
637 | dy11 = _mm_sub_ps(iy1,jy1); |
638 | dz11 = _mm_sub_ps(iz1,jz1); |
639 | dx12 = _mm_sub_ps(ix1,jx2); |
640 | dy12 = _mm_sub_ps(iy1,jy2); |
641 | dz12 = _mm_sub_ps(iz1,jz2); |
642 | dx20 = _mm_sub_ps(ix2,jx0); |
643 | dy20 = _mm_sub_ps(iy2,jy0); |
644 | dz20 = _mm_sub_ps(iz2,jz0); |
645 | dx21 = _mm_sub_ps(ix2,jx1); |
646 | dy21 = _mm_sub_ps(iy2,jy1); |
647 | dz21 = _mm_sub_ps(iz2,jz1); |
648 | dx22 = _mm_sub_ps(ix2,jx2); |
649 | dy22 = _mm_sub_ps(iy2,jy2); |
650 | dz22 = _mm_sub_ps(iz2,jz2); |
651 | |
652 | /* Calculate squared distance and things based on it */ |
653 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
654 | rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01); |
655 | rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02); |
656 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
657 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
658 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
659 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
660 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
661 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
662 | |
663 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
664 | rinv01 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq01); |
665 | rinv02 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq02); |
666 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
667 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
668 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
669 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
670 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
671 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
672 | |
673 | rinvsq00 = _mm_mul_ps(rinv00,rinv00); |
674 | rinvsq01 = _mm_mul_ps(rinv01,rinv01); |
675 | rinvsq02 = _mm_mul_ps(rinv02,rinv02); |
676 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
677 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
678 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
679 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
680 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
681 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
682 | |
683 | fjx0 = _mm_setzero_ps(); |
684 | fjy0 = _mm_setzero_ps(); |
685 | fjz0 = _mm_setzero_ps(); |
686 | fjx1 = _mm_setzero_ps(); |
687 | fjy1 = _mm_setzero_ps(); |
688 | fjz1 = _mm_setzero_ps(); |
689 | fjx2 = _mm_setzero_ps(); |
690 | fjy2 = _mm_setzero_ps(); |
691 | fjz2 = _mm_setzero_ps(); |
692 | |
693 | /************************** |
694 | * CALCULATE INTERACTIONS * |
695 | **************************/ |
696 | |
697 | r00 = _mm_mul_ps(rsq00,rinv00); |
698 | r00 = _mm_andnot_ps(dummy_mask,r00); |
699 | |
700 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
701 | rt = _mm_mul_ps(r00,vftabscale); |
702 | vfitab = _mm_cvttps_epi32(rt); |
703 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
704 | vfitab = _mm_slli_epi32(vfitab,3); |
705 | |
706 | /* REACTION-FIELD ELECTROSTATICS */ |
707 | velec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf)); |
708 | felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2)); |
709 | |
710 | /* CUBIC SPLINE TABLE DISPERSION */ |
711 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
712 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
713 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
714 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
715 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
716 | Heps = _mm_mul_ps(vfeps,H); |
717 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
718 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
719 | vvdw6 = _mm_mul_ps(c6_00,VV); |
720 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
721 | fvdw6 = _mm_mul_ps(c6_00,FF); |
722 | |
723 | /* CUBIC SPLINE TABLE REPULSION */ |
724 | vfitab = _mm_add_epi32(vfitab,ifour); |
725 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
726 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
727 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
728 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
729 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
730 | Heps = _mm_mul_ps(vfeps,H); |
731 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
732 | VV = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp)); |
733 | vvdw12 = _mm_mul_ps(c12_00,VV); |
734 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
735 | fvdw12 = _mm_mul_ps(c12_00,FF); |
736 | vvdw = _mm_add_ps(vvdw12,vvdw6); |
737 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
738 | |
739 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
740 | velec = _mm_andnot_ps(dummy_mask,velec); |
741 | velecsum = _mm_add_ps(velecsum,velec); |
742 | vvdw = _mm_andnot_ps(dummy_mask,vvdw); |
743 | vvdwsum = _mm_add_ps(vvdwsum,vvdw); |
744 | |
745 | fscal = _mm_add_ps(felec,fvdw); |
746 | |
747 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
748 | |
749 | /* Calculate temporary vectorial force */ |
750 | tx = _mm_mul_ps(fscal,dx00); |
751 | ty = _mm_mul_ps(fscal,dy00); |
752 | tz = _mm_mul_ps(fscal,dz00); |
753 | |
754 | /* Update vectorial force */ |
755 | fix0 = _mm_add_ps(fix0,tx); |
756 | fiy0 = _mm_add_ps(fiy0,ty); |
757 | fiz0 = _mm_add_ps(fiz0,tz); |
758 | |
759 | fjx0 = _mm_add_ps(fjx0,tx); |
760 | fjy0 = _mm_add_ps(fjy0,ty); |
761 | fjz0 = _mm_add_ps(fjz0,tz); |
762 | |
763 | /************************** |
764 | * CALCULATE INTERACTIONS * |
765 | **************************/ |
766 | |
767 | /* REACTION-FIELD ELECTROSTATICS */ |
768 | velec = _mm_mul_ps(qq01,_mm_sub_ps(_mm_add_ps(rinv01,_mm_mul_ps(krf,rsq01)),crf)); |
769 | felec = _mm_mul_ps(qq01,_mm_sub_ps(_mm_mul_ps(rinv01,rinvsq01),krf2)); |
770 | |
771 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
772 | velec = _mm_andnot_ps(dummy_mask,velec); |
773 | velecsum = _mm_add_ps(velecsum,velec); |
774 | |
775 | fscal = felec; |
776 | |
777 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
778 | |
779 | /* Calculate temporary vectorial force */ |
780 | tx = _mm_mul_ps(fscal,dx01); |
781 | ty = _mm_mul_ps(fscal,dy01); |
782 | tz = _mm_mul_ps(fscal,dz01); |
783 | |
784 | /* Update vectorial force */ |
785 | fix0 = _mm_add_ps(fix0,tx); |
786 | fiy0 = _mm_add_ps(fiy0,ty); |
787 | fiz0 = _mm_add_ps(fiz0,tz); |
788 | |
789 | fjx1 = _mm_add_ps(fjx1,tx); |
790 | fjy1 = _mm_add_ps(fjy1,ty); |
791 | fjz1 = _mm_add_ps(fjz1,tz); |
792 | |
793 | /************************** |
794 | * CALCULATE INTERACTIONS * |
795 | **************************/ |
796 | |
797 | /* REACTION-FIELD ELECTROSTATICS */ |
798 | velec = _mm_mul_ps(qq02,_mm_sub_ps(_mm_add_ps(rinv02,_mm_mul_ps(krf,rsq02)),crf)); |
799 | felec = _mm_mul_ps(qq02,_mm_sub_ps(_mm_mul_ps(rinv02,rinvsq02),krf2)); |
800 | |
801 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
802 | velec = _mm_andnot_ps(dummy_mask,velec); |
803 | velecsum = _mm_add_ps(velecsum,velec); |
804 | |
805 | fscal = felec; |
806 | |
807 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
808 | |
809 | /* Calculate temporary vectorial force */ |
810 | tx = _mm_mul_ps(fscal,dx02); |
811 | ty = _mm_mul_ps(fscal,dy02); |
812 | tz = _mm_mul_ps(fscal,dz02); |
813 | |
814 | /* Update vectorial force */ |
815 | fix0 = _mm_add_ps(fix0,tx); |
816 | fiy0 = _mm_add_ps(fiy0,ty); |
817 | fiz0 = _mm_add_ps(fiz0,tz); |
818 | |
819 | fjx2 = _mm_add_ps(fjx2,tx); |
820 | fjy2 = _mm_add_ps(fjy2,ty); |
821 | fjz2 = _mm_add_ps(fjz2,tz); |
822 | |
823 | /************************** |
824 | * CALCULATE INTERACTIONS * |
825 | **************************/ |
826 | |
827 | /* REACTION-FIELD ELECTROSTATICS */ |
828 | velec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf)); |
829 | felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2)); |
830 | |
831 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
832 | velec = _mm_andnot_ps(dummy_mask,velec); |
833 | velecsum = _mm_add_ps(velecsum,velec); |
834 | |
835 | fscal = felec; |
836 | |
837 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
838 | |
839 | /* Calculate temporary vectorial force */ |
840 | tx = _mm_mul_ps(fscal,dx10); |
841 | ty = _mm_mul_ps(fscal,dy10); |
842 | tz = _mm_mul_ps(fscal,dz10); |
843 | |
844 | /* Update vectorial force */ |
845 | fix1 = _mm_add_ps(fix1,tx); |
846 | fiy1 = _mm_add_ps(fiy1,ty); |
847 | fiz1 = _mm_add_ps(fiz1,tz); |
848 | |
849 | fjx0 = _mm_add_ps(fjx0,tx); |
850 | fjy0 = _mm_add_ps(fjy0,ty); |
851 | fjz0 = _mm_add_ps(fjz0,tz); |
852 | |
853 | /************************** |
854 | * CALCULATE INTERACTIONS * |
855 | **************************/ |
856 | |
857 | /* REACTION-FIELD ELECTROSTATICS */ |
858 | velec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_add_ps(rinv11,_mm_mul_ps(krf,rsq11)),crf)); |
859 | felec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_mul_ps(rinv11,rinvsq11),krf2)); |
860 | |
861 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
862 | velec = _mm_andnot_ps(dummy_mask,velec); |
863 | velecsum = _mm_add_ps(velecsum,velec); |
864 | |
865 | fscal = felec; |
866 | |
867 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
868 | |
869 | /* Calculate temporary vectorial force */ |
870 | tx = _mm_mul_ps(fscal,dx11); |
871 | ty = _mm_mul_ps(fscal,dy11); |
872 | tz = _mm_mul_ps(fscal,dz11); |
873 | |
874 | /* Update vectorial force */ |
875 | fix1 = _mm_add_ps(fix1,tx); |
876 | fiy1 = _mm_add_ps(fiy1,ty); |
877 | fiz1 = _mm_add_ps(fiz1,tz); |
878 | |
879 | fjx1 = _mm_add_ps(fjx1,tx); |
880 | fjy1 = _mm_add_ps(fjy1,ty); |
881 | fjz1 = _mm_add_ps(fjz1,tz); |
882 | |
883 | /************************** |
884 | * CALCULATE INTERACTIONS * |
885 | **************************/ |
886 | |
887 | /* REACTION-FIELD ELECTROSTATICS */ |
888 | velec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_add_ps(rinv12,_mm_mul_ps(krf,rsq12)),crf)); |
889 | felec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_mul_ps(rinv12,rinvsq12),krf2)); |
890 | |
891 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
892 | velec = _mm_andnot_ps(dummy_mask,velec); |
893 | velecsum = _mm_add_ps(velecsum,velec); |
894 | |
895 | fscal = felec; |
896 | |
897 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
898 | |
899 | /* Calculate temporary vectorial force */ |
900 | tx = _mm_mul_ps(fscal,dx12); |
901 | ty = _mm_mul_ps(fscal,dy12); |
902 | tz = _mm_mul_ps(fscal,dz12); |
903 | |
904 | /* Update vectorial force */ |
905 | fix1 = _mm_add_ps(fix1,tx); |
906 | fiy1 = _mm_add_ps(fiy1,ty); |
907 | fiz1 = _mm_add_ps(fiz1,tz); |
908 | |
909 | fjx2 = _mm_add_ps(fjx2,tx); |
910 | fjy2 = _mm_add_ps(fjy2,ty); |
911 | fjz2 = _mm_add_ps(fjz2,tz); |
912 | |
913 | /************************** |
914 | * CALCULATE INTERACTIONS * |
915 | **************************/ |
916 | |
917 | /* REACTION-FIELD ELECTROSTATICS */ |
918 | velec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf)); |
919 | felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2)); |
920 | |
921 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
922 | velec = _mm_andnot_ps(dummy_mask,velec); |
923 | velecsum = _mm_add_ps(velecsum,velec); |
924 | |
925 | fscal = felec; |
926 | |
927 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
928 | |
929 | /* Calculate temporary vectorial force */ |
930 | tx = _mm_mul_ps(fscal,dx20); |
931 | ty = _mm_mul_ps(fscal,dy20); |
932 | tz = _mm_mul_ps(fscal,dz20); |
933 | |
934 | /* Update vectorial force */ |
935 | fix2 = _mm_add_ps(fix2,tx); |
936 | fiy2 = _mm_add_ps(fiy2,ty); |
937 | fiz2 = _mm_add_ps(fiz2,tz); |
938 | |
939 | fjx0 = _mm_add_ps(fjx0,tx); |
940 | fjy0 = _mm_add_ps(fjy0,ty); |
941 | fjz0 = _mm_add_ps(fjz0,tz); |
942 | |
943 | /************************** |
944 | * CALCULATE INTERACTIONS * |
945 | **************************/ |
946 | |
947 | /* REACTION-FIELD ELECTROSTATICS */ |
948 | velec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_add_ps(rinv21,_mm_mul_ps(krf,rsq21)),crf)); |
949 | felec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_mul_ps(rinv21,rinvsq21),krf2)); |
950 | |
951 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
952 | velec = _mm_andnot_ps(dummy_mask,velec); |
953 | velecsum = _mm_add_ps(velecsum,velec); |
954 | |
955 | fscal = felec; |
956 | |
957 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
958 | |
959 | /* Calculate temporary vectorial force */ |
960 | tx = _mm_mul_ps(fscal,dx21); |
961 | ty = _mm_mul_ps(fscal,dy21); |
962 | tz = _mm_mul_ps(fscal,dz21); |
963 | |
964 | /* Update vectorial force */ |
965 | fix2 = _mm_add_ps(fix2,tx); |
966 | fiy2 = _mm_add_ps(fiy2,ty); |
967 | fiz2 = _mm_add_ps(fiz2,tz); |
968 | |
969 | fjx1 = _mm_add_ps(fjx1,tx); |
970 | fjy1 = _mm_add_ps(fjy1,ty); |
971 | fjz1 = _mm_add_ps(fjz1,tz); |
972 | |
973 | /************************** |
974 | * CALCULATE INTERACTIONS * |
975 | **************************/ |
976 | |
977 | /* REACTION-FIELD ELECTROSTATICS */ |
978 | velec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_add_ps(rinv22,_mm_mul_ps(krf,rsq22)),crf)); |
979 | felec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_mul_ps(rinv22,rinvsq22),krf2)); |
980 | |
981 | /* Update potential sum for this i atom from the interaction with this j atom. */ |
982 | velec = _mm_andnot_ps(dummy_mask,velec); |
983 | velecsum = _mm_add_ps(velecsum,velec); |
984 | |
985 | fscal = felec; |
986 | |
987 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
988 | |
989 | /* Calculate temporary vectorial force */ |
990 | tx = _mm_mul_ps(fscal,dx22); |
991 | ty = _mm_mul_ps(fscal,dy22); |
992 | tz = _mm_mul_ps(fscal,dz22); |
993 | |
994 | /* Update vectorial force */ |
995 | fix2 = _mm_add_ps(fix2,tx); |
996 | fiy2 = _mm_add_ps(fiy2,ty); |
997 | fiz2 = _mm_add_ps(fiz2,tz); |
998 | |
999 | fjx2 = _mm_add_ps(fjx2,tx); |
1000 | fjy2 = _mm_add_ps(fjy2,ty); |
1001 | fjz2 = _mm_add_ps(fjz2,tz); |
1002 | |
1003 | fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch; |
1004 | fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch; |
1005 | fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch; |
1006 | fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch; |
1007 | |
1008 | gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD, |
1009 | fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2); |
1010 | |
1011 | /* Inner loop uses 324 flops */ |
1012 | } |
1013 | |
1014 | /* End of innermost loop */ |
1015 | |
1016 | gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2, |
1017 | f+i_coord_offset,fshift+i_shift_offset); |
1018 | |
1019 | ggid = gid[iidx]; |
1020 | /* Update potential energies */ |
1021 | gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid); |
1022 | gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid); |
1023 | |
1024 | /* Increment number of inner iterations */ |
1025 | inneriter += j_index_end - j_index_start; |
1026 | |
1027 | /* Outer loop uses 20 flops */ |
1028 | } |
1029 | |
1030 | /* Increment number of outer iterations */ |
1031 | outeriter += nri; |
1032 | |
1033 | /* Update outer/inner flops */ |
1034 | |
1035 | inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*324)(nrnb)->n[eNR_NBKERNEL_ELEC_VDW_W3W3_VF] += outeriter*20 + inneriter*324; |
1036 | } |
1037 | /* |
1038 | * Gromacs nonbonded kernel: nb_kernel_ElecRF_VdwCSTab_GeomW3W3_F_sse4_1_single |
1039 | * Electrostatics interaction: ReactionField |
1040 | * VdW interaction: CubicSplineTable |
1041 | * Geometry: Water3-Water3 |
1042 | * Calculate force/pot: Force |
1043 | */ |
1044 | void |
1045 | nb_kernel_ElecRF_VdwCSTab_GeomW3W3_F_sse4_1_single |
1046 | (t_nblist * gmx_restrict nlist, |
1047 | rvec * gmx_restrict xx, |
1048 | rvec * gmx_restrict ff, |
1049 | t_forcerec * gmx_restrict fr, |
1050 | t_mdatoms * gmx_restrict mdatoms, |
1051 | nb_kernel_data_t gmx_unused__attribute__ ((unused)) * gmx_restrict kernel_data, |
1052 | t_nrnb * gmx_restrict nrnb) |
1053 | { |
1054 | /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or |
1055 | * just 0 for non-waters. |
1056 | * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different |
1057 | * jnr indices corresponding to data put in the four positions in the SIMD register. |
1058 | */ |
1059 | int i_shift_offset,i_coord_offset,outeriter,inneriter; |
1060 | int j_index_start,j_index_end,jidx,nri,inr,ggid,iidx; |
1061 | int jnrA,jnrB,jnrC,jnrD; |
1062 | int jnrlistA,jnrlistB,jnrlistC,jnrlistD; |
1063 | int j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD; |
1064 | int *iinr,*jindex,*jjnr,*shiftidx,*gid; |
1065 | real rcutoff_scalar; |
1066 | real *shiftvec,*fshift,*x,*f; |
1067 | real *fjptrA,*fjptrB,*fjptrC,*fjptrD; |
1068 | real scratch[4*DIM3]; |
1069 | __m128 tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall; |
1070 | int vdwioffset0; |
1071 | __m128 ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0; |
1072 | int vdwioffset1; |
1073 | __m128 ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1; |
1074 | int vdwioffset2; |
1075 | __m128 ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2; |
1076 | int vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D; |
1077 | __m128 jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0; |
1078 | int vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D; |
1079 | __m128 jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1; |
1080 | int vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D; |
1081 | __m128 jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2; |
1082 | __m128 dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00; |
1083 | __m128 dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01; |
1084 | __m128 dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02; |
1085 | __m128 dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10; |
1086 | __m128 dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11; |
1087 | __m128 dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12; |
1088 | __m128 dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20; |
1089 | __m128 dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21; |
1090 | __m128 dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22; |
1091 | __m128 velec,felec,velecsum,facel,crf,krf,krf2; |
1092 | real *charge; |
1093 | int nvdwtype; |
1094 | __m128 rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6; |
1095 | int *vdwtype; |
1096 | real *vdwparam; |
1097 | __m128 one_sixth = _mm_set1_ps(1.0/6.0); |
1098 | __m128 one_twelfth = _mm_set1_ps(1.0/12.0); |
1099 | __m128i vfitab; |
1100 | __m128i ifour = _mm_set1_epi32(4); |
1101 | __m128 rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF; |
1102 | real *vftab; |
1103 | __m128 dummy_mask,cutoff_mask; |
1104 | __m128 signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) ); |
1105 | __m128 one = _mm_set1_ps(1.0); |
1106 | __m128 two = _mm_set1_ps(2.0); |
1107 | x = xx[0]; |
1108 | f = ff[0]; |
1109 | |
1110 | nri = nlist->nri; |
1111 | iinr = nlist->iinr; |
1112 | jindex = nlist->jindex; |
1113 | jjnr = nlist->jjnr; |
1114 | shiftidx = nlist->shift; |
1115 | gid = nlist->gid; |
1116 | shiftvec = fr->shift_vec[0]; |
1117 | fshift = fr->fshift[0]; |
1118 | facel = _mm_set1_ps(fr->epsfac); |
1119 | charge = mdatoms->chargeA; |
1120 | krf = _mm_set1_ps(fr->ic->k_rf); |
1121 | krf2 = _mm_set1_ps(fr->ic->k_rf*2.0); |
1122 | crf = _mm_set1_ps(fr->ic->c_rf); |
Value stored to 'crf' is never read | |
1123 | nvdwtype = fr->ntype; |
1124 | vdwparam = fr->nbfp; |
1125 | vdwtype = mdatoms->typeA; |
1126 | |
1127 | vftab = kernel_data->table_vdw->data; |
1128 | vftabscale = _mm_set1_ps(kernel_data->table_vdw->scale); |
1129 | |
1130 | /* Setup water-specific parameters */ |
1131 | inr = nlist->iinr[0]; |
1132 | iq0 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0])); |
1133 | iq1 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1])); |
1134 | iq2 = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2])); |
1135 | vdwioffset0 = 2*nvdwtype*vdwtype[inr+0]; |
1136 | |
1137 | jq0 = _mm_set1_ps(charge[inr+0]); |
1138 | jq1 = _mm_set1_ps(charge[inr+1]); |
1139 | jq2 = _mm_set1_ps(charge[inr+2]); |
1140 | vdwjidx0A = 2*vdwtype[inr+0]; |
1141 | qq00 = _mm_mul_ps(iq0,jq0); |
1142 | c6_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]); |
1143 | c12_00 = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]); |
1144 | qq01 = _mm_mul_ps(iq0,jq1); |
1145 | qq02 = _mm_mul_ps(iq0,jq2); |
1146 | qq10 = _mm_mul_ps(iq1,jq0); |
1147 | qq11 = _mm_mul_ps(iq1,jq1); |
1148 | qq12 = _mm_mul_ps(iq1,jq2); |
1149 | qq20 = _mm_mul_ps(iq2,jq0); |
1150 | qq21 = _mm_mul_ps(iq2,jq1); |
1151 | qq22 = _mm_mul_ps(iq2,jq2); |
1152 | |
1153 | /* Avoid stupid compiler warnings */ |
1154 | jnrA = jnrB = jnrC = jnrD = 0; |
1155 | j_coord_offsetA = 0; |
1156 | j_coord_offsetB = 0; |
1157 | j_coord_offsetC = 0; |
1158 | j_coord_offsetD = 0; |
1159 | |
1160 | outeriter = 0; |
1161 | inneriter = 0; |
1162 | |
1163 | for(iidx=0;iidx<4*DIM3;iidx++) |
1164 | { |
1165 | scratch[iidx] = 0.0; |
1166 | } |
1167 | |
1168 | /* Start outer loop over neighborlists */ |
1169 | for(iidx=0; iidx<nri; iidx++) |
1170 | { |
1171 | /* Load shift vector for this list */ |
1172 | i_shift_offset = DIM3*shiftidx[iidx]; |
1173 | |
1174 | /* Load limits for loop over neighbors */ |
1175 | j_index_start = jindex[iidx]; |
1176 | j_index_end = jindex[iidx+1]; |
1177 | |
1178 | /* Get outer coordinate index */ |
1179 | inr = iinr[iidx]; |
1180 | i_coord_offset = DIM3*inr; |
1181 | |
1182 | /* Load i particle coords and add shift vector */ |
1183 | gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset, |
1184 | &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2); |
1185 | |
1186 | fix0 = _mm_setzero_ps(); |
1187 | fiy0 = _mm_setzero_ps(); |
1188 | fiz0 = _mm_setzero_ps(); |
1189 | fix1 = _mm_setzero_ps(); |
1190 | fiy1 = _mm_setzero_ps(); |
1191 | fiz1 = _mm_setzero_ps(); |
1192 | fix2 = _mm_setzero_ps(); |
1193 | fiy2 = _mm_setzero_ps(); |
1194 | fiz2 = _mm_setzero_ps(); |
1195 | |
1196 | /* Start inner kernel loop */ |
1197 | for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4) |
1198 | { |
1199 | |
1200 | /* Get j neighbor index, and coordinate index */ |
1201 | jnrA = jjnr[jidx]; |
1202 | jnrB = jjnr[jidx+1]; |
1203 | jnrC = jjnr[jidx+2]; |
1204 | jnrD = jjnr[jidx+3]; |
1205 | j_coord_offsetA = DIM3*jnrA; |
1206 | j_coord_offsetB = DIM3*jnrB; |
1207 | j_coord_offsetC = DIM3*jnrC; |
1208 | j_coord_offsetD = DIM3*jnrD; |
1209 | |
1210 | /* load j atom coordinates */ |
1211 | gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
1212 | x+j_coord_offsetC,x+j_coord_offsetD, |
1213 | &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2); |
1214 | |
1215 | /* Calculate displacement vector */ |
1216 | dx00 = _mm_sub_ps(ix0,jx0); |
1217 | dy00 = _mm_sub_ps(iy0,jy0); |
1218 | dz00 = _mm_sub_ps(iz0,jz0); |
1219 | dx01 = _mm_sub_ps(ix0,jx1); |
1220 | dy01 = _mm_sub_ps(iy0,jy1); |
1221 | dz01 = _mm_sub_ps(iz0,jz1); |
1222 | dx02 = _mm_sub_ps(ix0,jx2); |
1223 | dy02 = _mm_sub_ps(iy0,jy2); |
1224 | dz02 = _mm_sub_ps(iz0,jz2); |
1225 | dx10 = _mm_sub_ps(ix1,jx0); |
1226 | dy10 = _mm_sub_ps(iy1,jy0); |
1227 | dz10 = _mm_sub_ps(iz1,jz0); |
1228 | dx11 = _mm_sub_ps(ix1,jx1); |
1229 | dy11 = _mm_sub_ps(iy1,jy1); |
1230 | dz11 = _mm_sub_ps(iz1,jz1); |
1231 | dx12 = _mm_sub_ps(ix1,jx2); |
1232 | dy12 = _mm_sub_ps(iy1,jy2); |
1233 | dz12 = _mm_sub_ps(iz1,jz2); |
1234 | dx20 = _mm_sub_ps(ix2,jx0); |
1235 | dy20 = _mm_sub_ps(iy2,jy0); |
1236 | dz20 = _mm_sub_ps(iz2,jz0); |
1237 | dx21 = _mm_sub_ps(ix2,jx1); |
1238 | dy21 = _mm_sub_ps(iy2,jy1); |
1239 | dz21 = _mm_sub_ps(iz2,jz1); |
1240 | dx22 = _mm_sub_ps(ix2,jx2); |
1241 | dy22 = _mm_sub_ps(iy2,jy2); |
1242 | dz22 = _mm_sub_ps(iz2,jz2); |
1243 | |
1244 | /* Calculate squared distance and things based on it */ |
1245 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
1246 | rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01); |
1247 | rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02); |
1248 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
1249 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
1250 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
1251 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
1252 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
1253 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
1254 | |
1255 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
1256 | rinv01 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq01); |
1257 | rinv02 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq02); |
1258 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
1259 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
1260 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
1261 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
1262 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
1263 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
1264 | |
1265 | rinvsq00 = _mm_mul_ps(rinv00,rinv00); |
1266 | rinvsq01 = _mm_mul_ps(rinv01,rinv01); |
1267 | rinvsq02 = _mm_mul_ps(rinv02,rinv02); |
1268 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
1269 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
1270 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
1271 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
1272 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
1273 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
1274 | |
1275 | fjx0 = _mm_setzero_ps(); |
1276 | fjy0 = _mm_setzero_ps(); |
1277 | fjz0 = _mm_setzero_ps(); |
1278 | fjx1 = _mm_setzero_ps(); |
1279 | fjy1 = _mm_setzero_ps(); |
1280 | fjz1 = _mm_setzero_ps(); |
1281 | fjx2 = _mm_setzero_ps(); |
1282 | fjy2 = _mm_setzero_ps(); |
1283 | fjz2 = _mm_setzero_ps(); |
1284 | |
1285 | /************************** |
1286 | * CALCULATE INTERACTIONS * |
1287 | **************************/ |
1288 | |
1289 | r00 = _mm_mul_ps(rsq00,rinv00); |
1290 | |
1291 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
1292 | rt = _mm_mul_ps(r00,vftabscale); |
1293 | vfitab = _mm_cvttps_epi32(rt); |
1294 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
1295 | vfitab = _mm_slli_epi32(vfitab,3); |
1296 | |
1297 | /* REACTION-FIELD ELECTROSTATICS */ |
1298 | felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2)); |
1299 | |
1300 | /* CUBIC SPLINE TABLE DISPERSION */ |
1301 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
1302 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
1303 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
1304 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
1305 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
1306 | Heps = _mm_mul_ps(vfeps,H); |
1307 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
1308 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
1309 | fvdw6 = _mm_mul_ps(c6_00,FF); |
1310 | |
1311 | /* CUBIC SPLINE TABLE REPULSION */ |
1312 | vfitab = _mm_add_epi32(vfitab,ifour); |
1313 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
1314 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
1315 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
1316 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
1317 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
1318 | Heps = _mm_mul_ps(vfeps,H); |
1319 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
1320 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
1321 | fvdw12 = _mm_mul_ps(c12_00,FF); |
1322 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
1323 | |
1324 | fscal = _mm_add_ps(felec,fvdw); |
1325 | |
1326 | /* Calculate temporary vectorial force */ |
1327 | tx = _mm_mul_ps(fscal,dx00); |
1328 | ty = _mm_mul_ps(fscal,dy00); |
1329 | tz = _mm_mul_ps(fscal,dz00); |
1330 | |
1331 | /* Update vectorial force */ |
1332 | fix0 = _mm_add_ps(fix0,tx); |
1333 | fiy0 = _mm_add_ps(fiy0,ty); |
1334 | fiz0 = _mm_add_ps(fiz0,tz); |
1335 | |
1336 | fjx0 = _mm_add_ps(fjx0,tx); |
1337 | fjy0 = _mm_add_ps(fjy0,ty); |
1338 | fjz0 = _mm_add_ps(fjz0,tz); |
1339 | |
1340 | /************************** |
1341 | * CALCULATE INTERACTIONS * |
1342 | **************************/ |
1343 | |
1344 | /* REACTION-FIELD ELECTROSTATICS */ |
1345 | felec = _mm_mul_ps(qq01,_mm_sub_ps(_mm_mul_ps(rinv01,rinvsq01),krf2)); |
1346 | |
1347 | fscal = felec; |
1348 | |
1349 | /* Calculate temporary vectorial force */ |
1350 | tx = _mm_mul_ps(fscal,dx01); |
1351 | ty = _mm_mul_ps(fscal,dy01); |
1352 | tz = _mm_mul_ps(fscal,dz01); |
1353 | |
1354 | /* Update vectorial force */ |
1355 | fix0 = _mm_add_ps(fix0,tx); |
1356 | fiy0 = _mm_add_ps(fiy0,ty); |
1357 | fiz0 = _mm_add_ps(fiz0,tz); |
1358 | |
1359 | fjx1 = _mm_add_ps(fjx1,tx); |
1360 | fjy1 = _mm_add_ps(fjy1,ty); |
1361 | fjz1 = _mm_add_ps(fjz1,tz); |
1362 | |
1363 | /************************** |
1364 | * CALCULATE INTERACTIONS * |
1365 | **************************/ |
1366 | |
1367 | /* REACTION-FIELD ELECTROSTATICS */ |
1368 | felec = _mm_mul_ps(qq02,_mm_sub_ps(_mm_mul_ps(rinv02,rinvsq02),krf2)); |
1369 | |
1370 | fscal = felec; |
1371 | |
1372 | /* Calculate temporary vectorial force */ |
1373 | tx = _mm_mul_ps(fscal,dx02); |
1374 | ty = _mm_mul_ps(fscal,dy02); |
1375 | tz = _mm_mul_ps(fscal,dz02); |
1376 | |
1377 | /* Update vectorial force */ |
1378 | fix0 = _mm_add_ps(fix0,tx); |
1379 | fiy0 = _mm_add_ps(fiy0,ty); |
1380 | fiz0 = _mm_add_ps(fiz0,tz); |
1381 | |
1382 | fjx2 = _mm_add_ps(fjx2,tx); |
1383 | fjy2 = _mm_add_ps(fjy2,ty); |
1384 | fjz2 = _mm_add_ps(fjz2,tz); |
1385 | |
1386 | /************************** |
1387 | * CALCULATE INTERACTIONS * |
1388 | **************************/ |
1389 | |
1390 | /* REACTION-FIELD ELECTROSTATICS */ |
1391 | felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2)); |
1392 | |
1393 | fscal = felec; |
1394 | |
1395 | /* Calculate temporary vectorial force */ |
1396 | tx = _mm_mul_ps(fscal,dx10); |
1397 | ty = _mm_mul_ps(fscal,dy10); |
1398 | tz = _mm_mul_ps(fscal,dz10); |
1399 | |
1400 | /* Update vectorial force */ |
1401 | fix1 = _mm_add_ps(fix1,tx); |
1402 | fiy1 = _mm_add_ps(fiy1,ty); |
1403 | fiz1 = _mm_add_ps(fiz1,tz); |
1404 | |
1405 | fjx0 = _mm_add_ps(fjx0,tx); |
1406 | fjy0 = _mm_add_ps(fjy0,ty); |
1407 | fjz0 = _mm_add_ps(fjz0,tz); |
1408 | |
1409 | /************************** |
1410 | * CALCULATE INTERACTIONS * |
1411 | **************************/ |
1412 | |
1413 | /* REACTION-FIELD ELECTROSTATICS */ |
1414 | felec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_mul_ps(rinv11,rinvsq11),krf2)); |
1415 | |
1416 | fscal = felec; |
1417 | |
1418 | /* Calculate temporary vectorial force */ |
1419 | tx = _mm_mul_ps(fscal,dx11); |
1420 | ty = _mm_mul_ps(fscal,dy11); |
1421 | tz = _mm_mul_ps(fscal,dz11); |
1422 | |
1423 | /* Update vectorial force */ |
1424 | fix1 = _mm_add_ps(fix1,tx); |
1425 | fiy1 = _mm_add_ps(fiy1,ty); |
1426 | fiz1 = _mm_add_ps(fiz1,tz); |
1427 | |
1428 | fjx1 = _mm_add_ps(fjx1,tx); |
1429 | fjy1 = _mm_add_ps(fjy1,ty); |
1430 | fjz1 = _mm_add_ps(fjz1,tz); |
1431 | |
1432 | /************************** |
1433 | * CALCULATE INTERACTIONS * |
1434 | **************************/ |
1435 | |
1436 | /* REACTION-FIELD ELECTROSTATICS */ |
1437 | felec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_mul_ps(rinv12,rinvsq12),krf2)); |
1438 | |
1439 | fscal = felec; |
1440 | |
1441 | /* Calculate temporary vectorial force */ |
1442 | tx = _mm_mul_ps(fscal,dx12); |
1443 | ty = _mm_mul_ps(fscal,dy12); |
1444 | tz = _mm_mul_ps(fscal,dz12); |
1445 | |
1446 | /* Update vectorial force */ |
1447 | fix1 = _mm_add_ps(fix1,tx); |
1448 | fiy1 = _mm_add_ps(fiy1,ty); |
1449 | fiz1 = _mm_add_ps(fiz1,tz); |
1450 | |
1451 | fjx2 = _mm_add_ps(fjx2,tx); |
1452 | fjy2 = _mm_add_ps(fjy2,ty); |
1453 | fjz2 = _mm_add_ps(fjz2,tz); |
1454 | |
1455 | /************************** |
1456 | * CALCULATE INTERACTIONS * |
1457 | **************************/ |
1458 | |
1459 | /* REACTION-FIELD ELECTROSTATICS */ |
1460 | felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2)); |
1461 | |
1462 | fscal = felec; |
1463 | |
1464 | /* Calculate temporary vectorial force */ |
1465 | tx = _mm_mul_ps(fscal,dx20); |
1466 | ty = _mm_mul_ps(fscal,dy20); |
1467 | tz = _mm_mul_ps(fscal,dz20); |
1468 | |
1469 | /* Update vectorial force */ |
1470 | fix2 = _mm_add_ps(fix2,tx); |
1471 | fiy2 = _mm_add_ps(fiy2,ty); |
1472 | fiz2 = _mm_add_ps(fiz2,tz); |
1473 | |
1474 | fjx0 = _mm_add_ps(fjx0,tx); |
1475 | fjy0 = _mm_add_ps(fjy0,ty); |
1476 | fjz0 = _mm_add_ps(fjz0,tz); |
1477 | |
1478 | /************************** |
1479 | * CALCULATE INTERACTIONS * |
1480 | **************************/ |
1481 | |
1482 | /* REACTION-FIELD ELECTROSTATICS */ |
1483 | felec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_mul_ps(rinv21,rinvsq21),krf2)); |
1484 | |
1485 | fscal = felec; |
1486 | |
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); |
1491 | |
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); |
1496 | |
1497 | fjx1 = _mm_add_ps(fjx1,tx); |
1498 | fjy1 = _mm_add_ps(fjy1,ty); |
1499 | fjz1 = _mm_add_ps(fjz1,tz); |
1500 | |
1501 | /************************** |
1502 | * CALCULATE INTERACTIONS * |
1503 | **************************/ |
1504 | |
1505 | /* REACTION-FIELD ELECTROSTATICS */ |
1506 | felec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_mul_ps(rinv22,rinvsq22),krf2)); |
1507 | |
1508 | fscal = felec; |
1509 | |
1510 | /* Calculate temporary vectorial force */ |
1511 | tx = _mm_mul_ps(fscal,dx22); |
1512 | ty = _mm_mul_ps(fscal,dy22); |
1513 | tz = _mm_mul_ps(fscal,dz22); |
1514 | |
1515 | /* Update vectorial force */ |
1516 | fix2 = _mm_add_ps(fix2,tx); |
1517 | fiy2 = _mm_add_ps(fiy2,ty); |
1518 | fiz2 = _mm_add_ps(fiz2,tz); |
1519 | |
1520 | fjx2 = _mm_add_ps(fjx2,tx); |
1521 | fjy2 = _mm_add_ps(fjy2,ty); |
1522 | fjz2 = _mm_add_ps(fjz2,tz); |
1523 | |
1524 | fjptrA = f+j_coord_offsetA; |
1525 | fjptrB = f+j_coord_offsetB; |
1526 | fjptrC = f+j_coord_offsetC; |
1527 | fjptrD = f+j_coord_offsetD; |
1528 | |
1529 | gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD, |
1530 | fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2); |
1531 | |
1532 | /* Inner loop uses 270 flops */ |
1533 | } |
1534 | |
1535 | if(jidx<j_index_end) |
1536 | { |
1537 | |
1538 | /* Get j neighbor index, and coordinate index */ |
1539 | jnrlistA = jjnr[jidx]; |
1540 | jnrlistB = jjnr[jidx+1]; |
1541 | jnrlistC = jjnr[jidx+2]; |
1542 | jnrlistD = jjnr[jidx+3]; |
1543 | /* Sign of each element will be negative for non-real atoms. |
1544 | * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones, |
1545 | * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries. |
1546 | */ |
1547 | dummy_mask = gmx_mm_castsi128_ps_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())); |
1548 | jnrA = (jnrlistA>=0) ? jnrlistA : 0; |
1549 | jnrB = (jnrlistB>=0) ? jnrlistB : 0; |
1550 | jnrC = (jnrlistC>=0) ? jnrlistC : 0; |
1551 | jnrD = (jnrlistD>=0) ? jnrlistD : 0; |
1552 | j_coord_offsetA = DIM3*jnrA; |
1553 | j_coord_offsetB = DIM3*jnrB; |
1554 | j_coord_offsetC = DIM3*jnrC; |
1555 | j_coord_offsetD = DIM3*jnrD; |
1556 | |
1557 | /* load j atom coordinates */ |
1558 | gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB, |
1559 | x+j_coord_offsetC,x+j_coord_offsetD, |
1560 | &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2); |
1561 | |
1562 | /* Calculate displacement vector */ |
1563 | dx00 = _mm_sub_ps(ix0,jx0); |
1564 | dy00 = _mm_sub_ps(iy0,jy0); |
1565 | dz00 = _mm_sub_ps(iz0,jz0); |
1566 | dx01 = _mm_sub_ps(ix0,jx1); |
1567 | dy01 = _mm_sub_ps(iy0,jy1); |
1568 | dz01 = _mm_sub_ps(iz0,jz1); |
1569 | dx02 = _mm_sub_ps(ix0,jx2); |
1570 | dy02 = _mm_sub_ps(iy0,jy2); |
1571 | dz02 = _mm_sub_ps(iz0,jz2); |
1572 | dx10 = _mm_sub_ps(ix1,jx0); |
1573 | dy10 = _mm_sub_ps(iy1,jy0); |
1574 | dz10 = _mm_sub_ps(iz1,jz0); |
1575 | dx11 = _mm_sub_ps(ix1,jx1); |
1576 | dy11 = _mm_sub_ps(iy1,jy1); |
1577 | dz11 = _mm_sub_ps(iz1,jz1); |
1578 | dx12 = _mm_sub_ps(ix1,jx2); |
1579 | dy12 = _mm_sub_ps(iy1,jy2); |
1580 | dz12 = _mm_sub_ps(iz1,jz2); |
1581 | dx20 = _mm_sub_ps(ix2,jx0); |
1582 | dy20 = _mm_sub_ps(iy2,jy0); |
1583 | dz20 = _mm_sub_ps(iz2,jz0); |
1584 | dx21 = _mm_sub_ps(ix2,jx1); |
1585 | dy21 = _mm_sub_ps(iy2,jy1); |
1586 | dz21 = _mm_sub_ps(iz2,jz1); |
1587 | dx22 = _mm_sub_ps(ix2,jx2); |
1588 | dy22 = _mm_sub_ps(iy2,jy2); |
1589 | dz22 = _mm_sub_ps(iz2,jz2); |
1590 | |
1591 | /* Calculate squared distance and things based on it */ |
1592 | rsq00 = gmx_mm_calc_rsq_ps(dx00,dy00,dz00); |
1593 | rsq01 = gmx_mm_calc_rsq_ps(dx01,dy01,dz01); |
1594 | rsq02 = gmx_mm_calc_rsq_ps(dx02,dy02,dz02); |
1595 | rsq10 = gmx_mm_calc_rsq_ps(dx10,dy10,dz10); |
1596 | rsq11 = gmx_mm_calc_rsq_ps(dx11,dy11,dz11); |
1597 | rsq12 = gmx_mm_calc_rsq_ps(dx12,dy12,dz12); |
1598 | rsq20 = gmx_mm_calc_rsq_ps(dx20,dy20,dz20); |
1599 | rsq21 = gmx_mm_calc_rsq_ps(dx21,dy21,dz21); |
1600 | rsq22 = gmx_mm_calc_rsq_ps(dx22,dy22,dz22); |
1601 | |
1602 | rinv00 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq00); |
1603 | rinv01 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq01); |
1604 | rinv02 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq02); |
1605 | rinv10 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq10); |
1606 | rinv11 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq11); |
1607 | rinv12 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq12); |
1608 | rinv20 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq20); |
1609 | rinv21 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq21); |
1610 | rinv22 = gmx_mm_invsqrt_psgmx_simd_invsqrt_f(rsq22); |
1611 | |
1612 | rinvsq00 = _mm_mul_ps(rinv00,rinv00); |
1613 | rinvsq01 = _mm_mul_ps(rinv01,rinv01); |
1614 | rinvsq02 = _mm_mul_ps(rinv02,rinv02); |
1615 | rinvsq10 = _mm_mul_ps(rinv10,rinv10); |
1616 | rinvsq11 = _mm_mul_ps(rinv11,rinv11); |
1617 | rinvsq12 = _mm_mul_ps(rinv12,rinv12); |
1618 | rinvsq20 = _mm_mul_ps(rinv20,rinv20); |
1619 | rinvsq21 = _mm_mul_ps(rinv21,rinv21); |
1620 | rinvsq22 = _mm_mul_ps(rinv22,rinv22); |
1621 | |
1622 | fjx0 = _mm_setzero_ps(); |
1623 | fjy0 = _mm_setzero_ps(); |
1624 | fjz0 = _mm_setzero_ps(); |
1625 | fjx1 = _mm_setzero_ps(); |
1626 | fjy1 = _mm_setzero_ps(); |
1627 | fjz1 = _mm_setzero_ps(); |
1628 | fjx2 = _mm_setzero_ps(); |
1629 | fjy2 = _mm_setzero_ps(); |
1630 | fjz2 = _mm_setzero_ps(); |
1631 | |
1632 | /************************** |
1633 | * CALCULATE INTERACTIONS * |
1634 | **************************/ |
1635 | |
1636 | r00 = _mm_mul_ps(rsq00,rinv00); |
1637 | r00 = _mm_andnot_ps(dummy_mask,r00); |
1638 | |
1639 | /* Calculate table index by multiplying r with table scale and truncate to integer */ |
1640 | rt = _mm_mul_ps(r00,vftabscale); |
1641 | vfitab = _mm_cvttps_epi32(rt); |
1642 | vfeps = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR)__extension__ ({ __m128 __X = (rt); (__m128) __builtin_ia32_roundps ((__v4sf)__X, ((0x00 | 0x01))); })); |
1643 | vfitab = _mm_slli_epi32(vfitab,3); |
1644 | |
1645 | /* REACTION-FIELD ELECTROSTATICS */ |
1646 | felec = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2)); |
1647 | |
1648 | /* CUBIC SPLINE TABLE DISPERSION */ |
1649 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
1650 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
1651 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
1652 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
1653 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
1654 | Heps = _mm_mul_ps(vfeps,H); |
1655 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
1656 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
1657 | fvdw6 = _mm_mul_ps(c6_00,FF); |
1658 | |
1659 | /* CUBIC SPLINE TABLE REPULSION */ |
1660 | vfitab = _mm_add_epi32(vfitab,ifour); |
1661 | Y = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(0) & 3];})) ); |
1662 | F = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(1) & 3];})) ); |
1663 | G = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(2) & 3];})) ); |
1664 | H = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3)(__extension__ ({ __v4si __a = (__v4si)(vfitab); __a[(3) & 3];})) ); |
1665 | _MM_TRANSPOSE4_PS(Y,F,G,H)do { __m128 tmp3, tmp2, tmp1, tmp0; tmp0 = _mm_unpacklo_ps((Y ), (F)); tmp2 = _mm_unpacklo_ps((G), (H)); tmp1 = _mm_unpackhi_ps ((Y), (F)); tmp3 = _mm_unpackhi_ps((G), (H)); (Y) = _mm_movelh_ps (tmp0, tmp2); (F) = _mm_movehl_ps(tmp2, tmp0); (G) = _mm_movelh_ps (tmp1, tmp3); (H) = _mm_movehl_ps(tmp3, tmp1); } while (0); |
1666 | Heps = _mm_mul_ps(vfeps,H); |
1667 | Fp = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps))); |
1668 | FF = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps)))); |
1669 | fvdw12 = _mm_mul_ps(c12_00,FF); |
1670 | fvdw = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00))); |
1671 | |
1672 | fscal = _mm_add_ps(felec,fvdw); |
1673 | |
1674 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1675 | |
1676 | /* Calculate temporary vectorial force */ |
1677 | tx = _mm_mul_ps(fscal,dx00); |
1678 | ty = _mm_mul_ps(fscal,dy00); |
1679 | tz = _mm_mul_ps(fscal,dz00); |
1680 | |
1681 | /* Update vectorial force */ |
1682 | fix0 = _mm_add_ps(fix0,tx); |
1683 | fiy0 = _mm_add_ps(fiy0,ty); |
1684 | fiz0 = _mm_add_ps(fiz0,tz); |
1685 | |
1686 | fjx0 = _mm_add_ps(fjx0,tx); |
1687 | fjy0 = _mm_add_ps(fjy0,ty); |
1688 | fjz0 = _mm_add_ps(fjz0,tz); |
1689 | |
1690 | /************************** |
1691 | * CALCULATE INTERACTIONS * |
1692 | **************************/ |
1693 | |
1694 | /* REACTION-FIELD ELECTROSTATICS */ |
1695 | felec = _mm_mul_ps(qq01,_mm_sub_ps(_mm_mul_ps(rinv01,rinvsq01),krf2)); |
1696 | |
1697 | fscal = felec; |
1698 | |
1699 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1700 | |
1701 | /* Calculate temporary vectorial force */ |
1702 | tx = _mm_mul_ps(fscal,dx01); |
1703 | ty = _mm_mul_ps(fscal,dy01); |
1704 | tz = _mm_mul_ps(fscal,dz01); |
1705 | |
1706 | /* Update vectorial force */ |
1707 | fix0 = _mm_add_ps(fix0,tx); |
1708 | fiy0 = _mm_add_ps(fiy0,ty); |
1709 | fiz0 = _mm_add_ps(fiz0,tz); |
1710 | |
1711 | fjx1 = _mm_add_ps(fjx1,tx); |
1712 | fjy1 = _mm_add_ps(fjy1,ty); |
1713 | fjz1 = _mm_add_ps(fjz1,tz); |
1714 | |
1715 | /************************** |
1716 | * CALCULATE INTERACTIONS * |
1717 | **************************/ |
1718 | |
1719 | /* REACTION-FIELD ELECTROSTATICS */ |
1720 | felec = _mm_mul_ps(qq02,_mm_sub_ps(_mm_mul_ps(rinv02,rinvsq02),krf2)); |
1721 | |
1722 | fscal = felec; |
1723 | |
1724 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1725 | |
1726 | /* Calculate temporary vectorial force */ |
1727 | tx = _mm_mul_ps(fscal,dx02); |
1728 | ty = _mm_mul_ps(fscal,dy02); |
1729 | tz = _mm_mul_ps(fscal,dz02); |
1730 | |
1731 | /* Update vectorial force */ |
1732 | fix0 = _mm_add_ps(fix0,tx); |
1733 | fiy0 = _mm_add_ps(fiy0,ty); |
1734 | fiz0 = _mm_add_ps(fiz0,tz); |
1735 | |
1736 | fjx2 = _mm_add_ps(fjx2,tx); |
1737 | fjy2 = _mm_add_ps(fjy2,ty); |
1738 | fjz2 = _mm_add_ps(fjz2,tz); |
1739 | |
1740 | /************************** |
1741 | * CALCULATE INTERACTIONS * |
1742 | **************************/ |
1743 | |
1744 | /* REACTION-FIELD ELECTROSTATICS */ |
1745 | felec = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2)); |
1746 | |
1747 | fscal = felec; |
1748 | |
1749 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1750 | |
1751 | /* Calculate temporary vectorial force */ |
1752 | tx = _mm_mul_ps(fscal,dx10); |
1753 | ty = _mm_mul_ps(fscal,dy10); |
1754 | tz = _mm_mul_ps(fscal,dz10); |
1755 | |
1756 | /* Update vectorial force */ |
1757 | fix1 = _mm_add_ps(fix1,tx); |
1758 | fiy1 = _mm_add_ps(fiy1,ty); |
1759 | fiz1 = _mm_add_ps(fiz1,tz); |
1760 | |
1761 | fjx0 = _mm_add_ps(fjx0,tx); |
1762 | fjy0 = _mm_add_ps(fjy0,ty); |
1763 | fjz0 = _mm_add_ps(fjz0,tz); |
1764 | |
1765 | /************************** |
1766 | * CALCULATE INTERACTIONS * |
1767 | **************************/ |
1768 | |
1769 | /* REACTION-FIELD ELECTROSTATICS */ |
1770 | felec = _mm_mul_ps(qq11,_mm_sub_ps(_mm_mul_ps(rinv11,rinvsq11),krf2)); |
1771 | |
1772 | fscal = felec; |
1773 | |
1774 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1775 | |
1776 | /* Calculate temporary vectorial force */ |
1777 | tx = _mm_mul_ps(fscal,dx11); |
1778 | ty = _mm_mul_ps(fscal,dy11); |
1779 | tz = _mm_mul_ps(fscal,dz11); |
1780 | |
1781 | /* Update vectorial force */ |
1782 | fix1 = _mm_add_ps(fix1,tx); |
1783 | fiy1 = _mm_add_ps(fiy1,ty); |
1784 | fiz1 = _mm_add_ps(fiz1,tz); |
1785 | |
1786 | fjx1 = _mm_add_ps(fjx1,tx); |
1787 | fjy1 = _mm_add_ps(fjy1,ty); |
1788 | fjz1 = _mm_add_ps(fjz1,tz); |
1789 | |
1790 | /************************** |
1791 | * CALCULATE INTERACTIONS * |
1792 | **************************/ |
1793 | |
1794 | /* REACTION-FIELD ELECTROSTATICS */ |
1795 | felec = _mm_mul_ps(qq12,_mm_sub_ps(_mm_mul_ps(rinv12,rinvsq12),krf2)); |
1796 | |
1797 | fscal = felec; |
1798 | |
1799 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1800 | |
1801 | /* Calculate temporary vectorial force */ |
1802 | tx = _mm_mul_ps(fscal,dx12); |
1803 | ty = _mm_mul_ps(fscal,dy12); |
1804 | tz = _mm_mul_ps(fscal,dz12); |
1805 | |
1806 | /* Update vectorial force */ |
1807 | fix1 = _mm_add_ps(fix1,tx); |
1808 | fiy1 = _mm_add_ps(fiy1,ty); |
1809 | fiz1 = _mm_add_ps(fiz1,tz); |
1810 | |
1811 | fjx2 = _mm_add_ps(fjx2,tx); |
1812 | fjy2 = _mm_add_ps(fjy2,ty); |
1813 | fjz2 = _mm_add_ps(fjz2,tz); |
1814 | |
1815 | /************************** |
1816 | * CALCULATE INTERACTIONS * |
1817 | **************************/ |
1818 | |
1819 | /* REACTION-FIELD ELECTROSTATICS */ |
1820 | felec = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2)); |
1821 | |
1822 | fscal = felec; |
1823 | |
1824 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1825 | |
1826 | /* Calculate temporary vectorial force */ |
1827 | tx = _mm_mul_ps(fscal,dx20); |
1828 | ty = _mm_mul_ps(fscal,dy20); |
1829 | tz = _mm_mul_ps(fscal,dz20); |
1830 | |
1831 | /* Update vectorial force */ |
1832 | fix2 = _mm_add_ps(fix2,tx); |
1833 | fiy2 = _mm_add_ps(fiy2,ty); |
1834 | fiz2 = _mm_add_ps(fiz2,tz); |
1835 | |
1836 | fjx0 = _mm_add_ps(fjx0,tx); |
1837 | fjy0 = _mm_add_ps(fjy0,ty); |
1838 | fjz0 = _mm_add_ps(fjz0,tz); |
1839 | |
1840 | /************************** |
1841 | * CALCULATE INTERACTIONS * |
1842 | **************************/ |
1843 | |
1844 | /* REACTION-FIELD ELECTROSTATICS */ |
1845 | felec = _mm_mul_ps(qq21,_mm_sub_ps(_mm_mul_ps(rinv21,rinvsq21),krf2)); |
1846 | |
1847 | fscal = felec; |
1848 | |
1849 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1850 | |
1851 | /* Calculate temporary vectorial force */ |
1852 | tx = _mm_mul_ps(fscal,dx21); |
1853 | ty = _mm_mul_ps(fscal,dy21); |
1854 | tz = _mm_mul_ps(fscal,dz21); |
1855 | |
1856 | /* Update vectorial force */ |
1857 | fix2 = _mm_add_ps(fix2,tx); |
1858 | fiy2 = _mm_add_ps(fiy2,ty); |
1859 | fiz2 = _mm_add_ps(fiz2,tz); |
1860 | |
1861 | fjx1 = _mm_add_ps(fjx1,tx); |
1862 | fjy1 = _mm_add_ps(fjy1,ty); |
1863 | fjz1 = _mm_add_ps(fjz1,tz); |
1864 | |
1865 | /************************** |
1866 | * CALCULATE INTERACTIONS * |
1867 | **************************/ |
1868 | |
1869 | /* REACTION-FIELD ELECTROSTATICS */ |
1870 | felec = _mm_mul_ps(qq22,_mm_sub_ps(_mm_mul_ps(rinv22,rinvsq22),krf2)); |
1871 | |
1872 | fscal = felec; |
1873 | |
1874 | fscal = _mm_andnot_ps(dummy_mask,fscal); |
1875 | |
1876 | /* Calculate temporary vectorial force */ |
1877 | tx = _mm_mul_ps(fscal,dx22); |
1878 | ty = _mm_mul_ps(fscal,dy22); |
1879 | tz = _mm_mul_ps(fscal,dz22); |
1880 | |
1881 | /* Update vectorial force */ |
1882 | fix2 = _mm_add_ps(fix2,tx); |
1883 | fiy2 = _mm_add_ps(fiy2,ty); |
1884 | fiz2 = _mm_add_ps(fiz2,tz); |
1885 | |
1886 | fjx2 = _mm_add_ps(fjx2,tx); |
1887 | fjy2 = _mm_add_ps(fjy2,ty); |
1888 | fjz2 = _mm_add_ps(fjz2,tz); |
1889 | |
1890 | fjptrA = (jnrlistA>=0) ? f+j_coord_offsetA : scratch; |
1891 | fjptrB = (jnrlistB>=0) ? f+j_coord_offsetB : scratch; |
1892 | fjptrC = (jnrlistC>=0) ? f+j_coord_offsetC : scratch; |
1893 | fjptrD = (jnrlistD>=0) ? f+j_coord_offsetD : scratch; |
1894 | |
1895 | gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD, |
1896 | fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2); |
1897 | |
1898 | /* Inner loop uses 271 flops */ |
1899 | } |
1900 | |
1901 | /* End of innermost loop */ |
1902 | |
1903 | gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2, |
1904 | f+i_coord_offset,fshift+i_shift_offset); |
1905 | |
1906 | /* Increment number of inner iterations */ |
1907 | inneriter += j_index_end - j_index_start; |
1908 | |
1909 | /* Outer loop uses 18 flops */ |
1910 | } |
1911 | |
1912 | /* Increment number of outer iterations */ |
1913 | outeriter += nri; |
1914 | |
1915 | /* Update outer/inner flops */ |
1916 | |
1917 | inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*271)(nrnb)->n[eNR_NBKERNEL_ELEC_VDW_W3W3_F] += outeriter*18 + inneriter *271; |
1918 | } |