src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/kernelutil_x86_sse4_1_double.h

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifndef _kernelutil_x86_sse4_1_double_h_
  36 #define _kernelutil_x86_sse4_1_double_h_
  37
  38 #include <math.h>
  39
  40 #include "gromacs/simd/general_x86_sse4_1.h"
  41
  42 #include <stdio.h>
  43
  44
  45 /* Normal sum of four ymm registers */
  46 #define gmx_mm_sum4_pd(t0, t1, t2, t3)  _mm_add_pd(_mm_add_pd(t0, t1), _mm_add_pd(t2, t3))
  47
  48 static int
  49 gmx_mm_any_lt(__m128d a, __m128d b)
  50 {
  51     return _mm_movemask_pd(_mm_cmplt_pd(a, b));
  52 }
  53
  54 static gmx_inline __m128d
  55 gmx_mm_calc_rsq_pd(__m128d dx, __m128d dy, __m128d dz)
  56 {
  57     return _mm_add_pd( _mm_add_pd( _mm_mul_pd(dx, dx), _mm_mul_pd(dy, dy) ), _mm_mul_pd(dz, dz) );
  58 }
  59
  60
  61 /* Load a double value from 1-2 places, merge into xmm register */
  62 static gmx_inline __m128d
  63 gmx_mm_load_2real_swizzle_pd(const double * gmx_restrict ptrA,
  64                              const double * gmx_restrict ptrB)
  65 {
  66     return _mm_unpacklo_pd(_mm_load_sd(ptrA), _mm_load_sd(ptrB));
  67 }
  68
  69 static gmx_inline __m128d
  70 gmx_mm_load_1real_pd(const double * gmx_restrict ptrA)
  71 {
  72     return _mm_load_sd(ptrA);
  73 }
  74
  75
  76 static gmx_inline void
  77 gmx_mm_store_2real_swizzle_pd(double * gmx_restrict ptrA,
  78                               double * gmx_restrict ptrB,
  79                               __m128d               xmm1)
  80 {
  81     __m128d t2;
  82
  83     t2       = _mm_unpackhi_pd(xmm1, xmm1);
  84     _mm_store_sd(ptrA, xmm1);
  85     _mm_store_sd(ptrB, t2);
  86 }
  87
  88 static gmx_inline void
  89 gmx_mm_store_1real_pd(double * gmx_restrict ptrA, __m128d xmm1)
  90 {
  91     _mm_store_sd(ptrA, xmm1);
  92 }
  93
  94
  95 /* Similar to store, but increments value in memory */
  96 static gmx_inline void
  97 gmx_mm_increment_2real_swizzle_pd(double * gmx_restrict ptrA,
  98                                   double * gmx_restrict ptrB, __m128d xmm1)
  99 {
 100     __m128d t1;
 101
 102     t1   = _mm_unpackhi_pd(xmm1, xmm1);
 103     xmm1 = _mm_add_sd(xmm1, _mm_load_sd(ptrA));
 104     t1   = _mm_add_sd(t1, _mm_load_sd(ptrB));
 105     _mm_store_sd(ptrA, xmm1);
 106     _mm_store_sd(ptrB, t1);
 107 }
 108
 109 static gmx_inline void
 110 gmx_mm_increment_1real_pd(double * gmx_restrict ptrA, __m128d xmm1)
 111 {
 112     __m128d tmp;
 113
 114     tmp = gmx_mm_load_1real_pd(ptrA);
 115     tmp = _mm_add_sd(tmp, xmm1);
 116     gmx_mm_store_1real_pd(ptrA, tmp);
 117 }
 118
 119
 120 static gmx_inline void
 121 gmx_mm_load_2pair_swizzle_pd(const double * gmx_restrict p1,
 122                              const double * gmx_restrict p2,
 123                              __m128d * gmx_restrict      c6,
 124                              __m128d * gmx_restrict      c12)
 125 {
 126     __m128d t1, t2, t3;
 127
 128     t1   = _mm_loadu_pd(p1);
 129     t2   = _mm_loadu_pd(p2);
 130     *c6  = _mm_unpacklo_pd(t1, t2);
 131     *c12 = _mm_unpackhi_pd(t1, t2);
 132 }
 133
 134 static gmx_inline void
 135 gmx_mm_load_1pair_swizzle_pd(const double * gmx_restrict p1,
 136                              __m128d * gmx_restrict      c6,
 137                              __m128d * gmx_restrict      c12)
 138 {
 139     *c6     = _mm_load_sd(p1);
 140     *c12    = _mm_load_sd(p1+1);
 141 }
 142
 143
 144
 145 static gmx_inline void
 146 gmx_mm_load_shift_and_1rvec_broadcast_pd(const double * gmx_restrict xyz_shift,
 147                                          const double * gmx_restrict xyz,
 148                                          __m128d * gmx_restrict      x1,
 149                                          __m128d * gmx_restrict      y1,
 150                                          __m128d * gmx_restrict      z1)
 151 {
 152     __m128d mem_xy, mem_z, mem_sxy, mem_sz;
 153
 154     mem_xy  = _mm_loadu_pd(xyz);
 155     mem_z   = _mm_load_sd(xyz+2);
 156     mem_sxy = _mm_loadu_pd(xyz_shift);
 157     mem_sz  = _mm_load_sd(xyz_shift+2);
 158
 159     mem_xy  = _mm_add_pd(mem_xy, mem_sxy);
 160     mem_z   = _mm_add_pd(mem_z, mem_sz);
 161
 162     *x1  = _mm_shuffle_pd(mem_xy, mem_xy, _MM_SHUFFLE2(0, 0));
 163     *y1  = _mm_shuffle_pd(mem_xy, mem_xy, _MM_SHUFFLE2(1, 1));
 164     *z1  = _mm_shuffle_pd(mem_z, mem_z, _MM_SHUFFLE2(0, 0));
 165 }
 166
 167
 168 static gmx_inline void
 169 gmx_mm_load_shift_and_3rvec_broadcast_pd(const double * gmx_restrict xyz_shift,
 170                                          const double * gmx_restrict xyz,
 171                                          __m128d * gmx_restrict x1, __m128d * gmx_restrict y1, __m128d * gmx_restrict z1,
 172                                          __m128d * gmx_restrict x2, __m128d * gmx_restrict y2, __m128d * gmx_restrict z2,
 173                                          __m128d * gmx_restrict x3, __m128d * gmx_restrict y3, __m128d * gmx_restrict z3)
 174 {
 175     __m128d t1, t2, t3, t4, t5, sxy, sz, szx, syz;
 176
 177     t1  = _mm_loadu_pd(xyz);
 178     t2  = _mm_loadu_pd(xyz+2);
 179     t3  = _mm_loadu_pd(xyz+4);
 180     t4  = _mm_loadu_pd(xyz+6);
 181     t5  = _mm_load_sd(xyz+8);
 182
 183     sxy = _mm_loadu_pd(xyz_shift);
 184     sz  = _mm_load_sd(xyz_shift+2);
 185     szx = _mm_shuffle_pd(sz, sxy, _MM_SHUFFLE2(0, 0));
 186     syz = _mm_shuffle_pd(sxy, sz, _MM_SHUFFLE2(0, 1));
 187
 188     t1  = _mm_add_pd(t1, sxy);
 189     t2  = _mm_add_pd(t2, szx);
 190     t3  = _mm_add_pd(t3, syz);
 191     t4  = _mm_add_pd(t4, sxy);
 192     t5  = _mm_add_sd(t5, sz);
 193
 194     *x1  = _mm_shuffle_pd(t1, t1, _MM_SHUFFLE2(0, 0));
 195     *y1  = _mm_shuffle_pd(t1, t1, _MM_SHUFFLE2(1, 1));
 196     *z1  = _mm_shuffle_pd(t2, t2, _MM_SHUFFLE2(0, 0));
 197     *x2  = _mm_shuffle_pd(t2, t2, _MM_SHUFFLE2(1, 1));
 198     *y2  = _mm_shuffle_pd(t3, t3, _MM_SHUFFLE2(0, 0));
 199     *z2  = _mm_shuffle_pd(t3, t3, _MM_SHUFFLE2(1, 1));
 200     *x3  = _mm_shuffle_pd(t4, t4, _MM_SHUFFLE2(0, 0));
 201     *y3  = _mm_shuffle_pd(t4, t4, _MM_SHUFFLE2(1, 1));
 202     *z3  = _mm_shuffle_pd(t5, t5, _MM_SHUFFLE2(0, 0));
 203 }
 204
 205
 206 static gmx_inline void
 207 gmx_mm_load_shift_and_4rvec_broadcast_pd(const double * gmx_restrict xyz_shift,
 208                                          const double * gmx_restrict xyz,
 209                                          __m128d * gmx_restrict x1, __m128d * gmx_restrict y1, __m128d * gmx_restrict z1,
 210                                          __m128d * gmx_restrict x2, __m128d * gmx_restrict y2, __m128d * gmx_restrict z2,
 211                                          __m128d * gmx_restrict x3, __m128d * gmx_restrict y3, __m128d * gmx_restrict z3,
 212                                          __m128d * gmx_restrict x4, __m128d * gmx_restrict y4, __m128d * gmx_restrict z4)
 213 {
 214     __m128d t1, t2, t3, t4, t5, t6, sxy, sz, szx, syz;
 215
 216     t1  = _mm_loadu_pd(xyz);
 217     t2  = _mm_loadu_pd(xyz+2);
 218     t3  = _mm_loadu_pd(xyz+4);
 219     t4  = _mm_loadu_pd(xyz+6);
 220     t5  = _mm_loadu_pd(xyz+8);
 221     t6  = _mm_loadu_pd(xyz+10);
 222
 223     sxy = _mm_loadu_pd(xyz_shift);
 224     sz  = _mm_load_sd(xyz_shift+2);
 225     szx = _mm_shuffle_pd(sz, sxy, _MM_SHUFFLE2(0, 0));
 226     syz = _mm_shuffle_pd(sxy, sz, _MM_SHUFFLE2(0, 1));
 227
 228     t1  = _mm_add_pd(t1, sxy);
 229     t2  = _mm_add_pd(t2, szx);
 230     t3  = _mm_add_pd(t3, syz);
 231     t4  = _mm_add_pd(t4, sxy);
 232     t5  = _mm_add_pd(t5, szx);
 233     t6  = _mm_add_pd(t6, syz);
 234
 235     *x1  = _mm_shuffle_pd(t1, t1, _MM_SHUFFLE2(0, 0));
 236     *y1  = _mm_shuffle_pd(t1, t1, _MM_SHUFFLE2(1, 1));
 237     *z1  = _mm_shuffle_pd(t2, t2, _MM_SHUFFLE2(0, 0));
 238     *x2  = _mm_shuffle_pd(t2, t2, _MM_SHUFFLE2(1, 1));
 239     *y2  = _mm_shuffle_pd(t3, t3, _MM_SHUFFLE2(0, 0));
 240     *z2  = _mm_shuffle_pd(t3, t3, _MM_SHUFFLE2(1, 1));
 241     *x3  = _mm_shuffle_pd(t4, t4, _MM_SHUFFLE2(0, 0));
 242     *y3  = _mm_shuffle_pd(t4, t4, _MM_SHUFFLE2(1, 1));
 243     *z3  = _mm_shuffle_pd(t5, t5, _MM_SHUFFLE2(0, 0));
 244     *x4  = _mm_shuffle_pd(t5, t5, _MM_SHUFFLE2(1, 1));
 245     *y4  = _mm_shuffle_pd(t6, t6, _MM_SHUFFLE2(0, 0));
 246     *z4  = _mm_shuffle_pd(t6, t6, _MM_SHUFFLE2(1, 1));
 247 }
 248
 249
 250
 251
 252 static gmx_inline void
 253 gmx_mm_load_1rvec_1ptr_swizzle_pd(const double * gmx_restrict p1,
 254                                   __m128d * gmx_restrict x, __m128d * gmx_restrict y, __m128d * gmx_restrict z)
 255 {
 256     *x            = _mm_load_sd(p1);
 257     *y            = _mm_load_sd(p1+1);
 258     *z            = _mm_load_sd(p1+2);
 259 }
 260
 261 static gmx_inline void
 262 gmx_mm_load_3rvec_1ptr_swizzle_pd(const double * gmx_restrict p1,
 263                                   __m128d * gmx_restrict x1, __m128d * gmx_restrict y1, __m128d * gmx_restrict z1,
 264                                   __m128d * gmx_restrict x2, __m128d * gmx_restrict y2, __m128d * gmx_restrict z2,
 265                                   __m128d * gmx_restrict x3, __m128d * gmx_restrict y3, __m128d * gmx_restrict z3)
 266 {
 267     *x1            = _mm_load_sd(p1);
 268     *y1            = _mm_load_sd(p1+1);
 269     *z1            = _mm_load_sd(p1+2);
 270     *x2            = _mm_load_sd(p1+3);
 271     *y2            = _mm_load_sd(p1+4);
 272     *z2            = _mm_load_sd(p1+5);
 273     *x3            = _mm_load_sd(p1+6);
 274     *y3            = _mm_load_sd(p1+7);
 275     *z3            = _mm_load_sd(p1+8);
 276 }
 277
 278 static gmx_inline void
 279 gmx_mm_load_4rvec_1ptr_swizzle_pd(const double * gmx_restrict p1,
 280                                   __m128d * gmx_restrict x1, __m128d * gmx_restrict y1, __m128d * gmx_restrict z1,
 281                                   __m128d * gmx_restrict x2, __m128d * gmx_restrict y2, __m128d * gmx_restrict z2,
 282                                   __m128d * gmx_restrict x3, __m128d * gmx_restrict y3, __m128d * gmx_restrict z3,
 283                                   __m128d * gmx_restrict x4, __m128d * gmx_restrict y4, __m128d * gmx_restrict z4)
 284 {
 285     *x1            = _mm_load_sd(p1);
 286     *y1            = _mm_load_sd(p1+1);
 287     *z1            = _mm_load_sd(p1+2);
 288     *x2            = _mm_load_sd(p1+3);
 289     *y2            = _mm_load_sd(p1+4);
 290     *z2            = _mm_load_sd(p1+5);
 291     *x3            = _mm_load_sd(p1+6);
 292     *y3            = _mm_load_sd(p1+7);
 293     *z3            = _mm_load_sd(p1+8);
 294     *x4            = _mm_load_sd(p1+9);
 295     *y4            = _mm_load_sd(p1+10);
 296     *z4            = _mm_load_sd(p1+11);
 297 }
 298
 299
 300 static gmx_inline void
 301 gmx_mm_load_1rvec_2ptr_swizzle_pd(const double * gmx_restrict ptrA,
 302                                   const double * gmx_restrict ptrB,
 303                                   __m128d * gmx_restrict x1, __m128d * gmx_restrict y1, __m128d * gmx_restrict z1)
 304 {
 305     __m128d t1, t2, t3, t4;
 306     t1           = _mm_loadu_pd(ptrA);
 307     t2           = _mm_loadu_pd(ptrB);
 308     t3           = _mm_load_sd(ptrA+2);
 309     t4           = _mm_load_sd(ptrB+2);
 310     GMX_MM_TRANSPOSE2_PD(t1, t2);
 311     *x1          = t1;
 312     *y1          = t2;
 313     *z1          = _mm_unpacklo_pd(t3, t4);
 314 }
 315
 316
 317 static gmx_inline void
 318 gmx_mm_load_3rvec_2ptr_swizzle_pd(const double * gmx_restrict ptrA, const double * gmx_restrict ptrB,
 319                                   __m128d * gmx_restrict x1, __m128d * gmx_restrict y1, __m128d * gmx_restrict z1,
 320                                   __m128d * gmx_restrict x2, __m128d * gmx_restrict y2, __m128d * gmx_restrict z2,
 321                                   __m128d * gmx_restrict x3, __m128d * gmx_restrict y3, __m128d * gmx_restrict z3)
 322 {
 323     __m128d t1, t2, t3, t4, t5, t6, t7, t8, t9, t10;
 324     t1           = _mm_loadu_pd(ptrA);
 325     t2           = _mm_loadu_pd(ptrB);
 326     t3           = _mm_loadu_pd(ptrA+2);
 327     t4           = _mm_loadu_pd(ptrB+2);
 328     t5           = _mm_loadu_pd(ptrA+4);
 329     t6           = _mm_loadu_pd(ptrB+4);
 330     t7           = _mm_loadu_pd(ptrA+6);
 331     t8           = _mm_loadu_pd(ptrB+6);
 332     t9           = _mm_load_sd(ptrA+8);
 333     t10          = _mm_load_sd(ptrB+8);
 334     GMX_MM_TRANSPOSE2_PD(t1, t2);
 335     GMX_MM_TRANSPOSE2_PD(t3, t4);
 336     GMX_MM_TRANSPOSE2_PD(t5, t6);
 337     GMX_MM_TRANSPOSE2_PD(t7, t8);
 338     *x1          = t1;
 339     *y1          = t2;
 340     *z1          = t3;
 341     *x2          = t4;
 342     *y2          = t5;
 343     *z2          = t6;
 344     *x3          = t7;
 345     *y3          = t8;
 346     *z3          = _mm_unpacklo_pd(t9, t10);
 347 }
 348
 349
 350 static gmx_inline void
 351 gmx_mm_load_4rvec_2ptr_swizzle_pd(const double * gmx_restrict ptrA, const double * gmx_restrict ptrB,
 352                                   __m128d * gmx_restrict x1, __m128d * gmx_restrict y1, __m128d * gmx_restrict z1,
 353                                   __m128d * gmx_restrict x2, __m128d * gmx_restrict y2, __m128d * gmx_restrict z2,
 354                                   __m128d * gmx_restrict x3, __m128d * gmx_restrict y3, __m128d * gmx_restrict z3,
 355                                   __m128d * gmx_restrict x4, __m128d * gmx_restrict y4, __m128d * gmx_restrict z4)
 356 {
 357     __m128d t1, t2, t3, t4, t5, t6;
 358     t1           = _mm_loadu_pd(ptrA);
 359     t2           = _mm_loadu_pd(ptrB);
 360     t3           = _mm_loadu_pd(ptrA+2);
 361     t4           = _mm_loadu_pd(ptrB+2);
 362     t5           = _mm_loadu_pd(ptrA+4);
 363     t6           = _mm_loadu_pd(ptrB+4);
 364     GMX_MM_TRANSPOSE2_PD(t1, t2);
 365     GMX_MM_TRANSPOSE2_PD(t3, t4);
 366     GMX_MM_TRANSPOSE2_PD(t5, t6);
 367     *x1          = t1;
 368     *y1          = t2;
 369     *z1          = t3;
 370     *x2          = t4;
 371     *y2          = t5;
 372     *z2          = t6;
 373     t1           = _mm_loadu_pd(ptrA+6);
 374     t2           = _mm_loadu_pd(ptrB+6);
 375     t3           = _mm_loadu_pd(ptrA+8);
 376     t4           = _mm_loadu_pd(ptrB+8);
 377     t5           = _mm_loadu_pd(ptrA+10);
 378     t6           = _mm_loadu_pd(ptrB+10);
 379     GMX_MM_TRANSPOSE2_PD(t1, t2);
 380     GMX_MM_TRANSPOSE2_PD(t3, t4);
 381     GMX_MM_TRANSPOSE2_PD(t5, t6);
 382     *x3          = t1;
 383     *y3          = t2;
 384     *z3          = t3;
 385     *x4          = t4;
 386     *y4          = t5;
 387     *z4          = t6;
 388 }
 389
 390
 391 /* Routines to decrement rvec in memory, typically use for j particle force updates */
 392 static gmx_inline void
 393 gmx_mm_decrement_1rvec_1ptr_noswizzle_pd(double * gmx_restrict ptrA,
 394                                          __m128d xy, __m128d z)
 395 {
 396     __m128d t1, t2;
 397
 398     t1 = _mm_loadu_pd(ptrA);
 399     t2 = _mm_load_sd(ptrA+2);
 400
 401     t1 = _mm_sub_pd(t1, xy);
 402     t2 = _mm_sub_sd(t2, z);
 403
 404     _mm_storeu_pd(ptrA, t1);
 405     _mm_store_sd(ptrA+2, t2);
 406 }
 407
 408
 409 static gmx_inline void
 410 gmx_mm_decrement_1rvec_1ptr_swizzle_pd(double * gmx_restrict ptrA,
 411                                        __m128d x1, __m128d y1, __m128d z1)
 412 {
 413     __m128d t1, t2, t3;
 414
 415     t1           = _mm_load_sd(ptrA);
 416     t2           = _mm_load_sd(ptrA+1);
 417     t3           = _mm_load_sd(ptrA+2);
 418
 419     t1           = _mm_sub_sd(t1, x1);
 420     t2           = _mm_sub_sd(t2, y1);
 421     t3           = _mm_sub_sd(t3, z1);
 422     _mm_store_sd(ptrA, t1);
 423     _mm_store_sd(ptrA+1, t2);
 424     _mm_store_sd(ptrA+2, t3);
 425 }
 426
 427
 428 #if defined (_MSC_VER) && defined(_M_IX86)
 429 /* Macro work-around since 32-bit MSVC cannot handle >3 xmm/ymm parameters */
 430 #define gmx_mm_decrement_3rvec_1ptr_swizzle_pd(ptrA, _x1, _y1, _z1, _x2, _y2, _z2, _x3, _y3, _z3) \
 431     { \
 432         __m128d _t1, _t2, _t3, _t4, _t5; \
 433         _t1          = _mm_loadu_pd(ptrA); \
 434         _t2          = _mm_loadu_pd(ptrA+2); \
 435         _t3          = _mm_loadu_pd(ptrA+4); \
 436         _t4          = _mm_loadu_pd(ptrA+6); \
 437         _t5          = _mm_load_sd(ptrA+8); \
 438         _x1          = _mm_unpacklo_pd(_x1, _y1); \
 439         _z1          = _mm_unpacklo_pd(_z1, _x2); \
 440         _y2          = _mm_unpacklo_pd(_y2, _z2); \
 441         _x3          = _mm_unpacklo_pd(_x3, _y3); \
 442         _t1          = _mm_sub_pd(_t1, _x1); \
 443         _t2          = _mm_sub_pd(_t2, _z1); \
 444         _t3          = _mm_sub_pd(_t3, _y2); \
 445         _t4          = _mm_sub_pd(_t4, _x3); \
 446         _t5          = _mm_sub_sd(_t5, _z3); \
 447         _mm_storeu_pd(ptrA, _t1); \
 448         _mm_storeu_pd(ptrA+2, _t2); \
 449         _mm_storeu_pd(ptrA+4, _t3); \
 450         _mm_storeu_pd(ptrA+6, _t4); \
 451         _mm_store_sd(ptrA+8, _t5); \
 452     }
 453 #else
 454 /* Real function for sane compilers */
 455 static gmx_inline void
 456 gmx_mm_decrement_3rvec_1ptr_swizzle_pd(double * gmx_restrict ptrA,
 457                                        __m128d x1, __m128d y1, __m128d z1,
 458                                        __m128d x2, __m128d y2, __m128d z2,
 459                                        __m128d x3, __m128d y3, __m128d z3)
 460 {
 461     __m128d t1, t2, t3, t4, t5;
 462
 463     t1          = _mm_loadu_pd(ptrA);
 464     t2          = _mm_loadu_pd(ptrA+2);
 465     t3          = _mm_loadu_pd(ptrA+4);
 466     t4          = _mm_loadu_pd(ptrA+6);
 467     t5          = _mm_load_sd(ptrA+8);
 468
 469     x1          = _mm_unpacklo_pd(x1, y1);
 470     z1          = _mm_unpacklo_pd(z1, x2);
 471     y2          = _mm_unpacklo_pd(y2, z2);
 472     x3          = _mm_unpacklo_pd(x3, y3);
 473     /* nothing to be done for z3 */
 474
 475     t1          = _mm_sub_pd(t1, x1);
 476     t2          = _mm_sub_pd(t2, z1);
 477     t3          = _mm_sub_pd(t3, y2);
 478     t4          = _mm_sub_pd(t4, x3);
 479     t5          = _mm_sub_sd(t5, z3);
 480     _mm_storeu_pd(ptrA, t1);
 481     _mm_storeu_pd(ptrA+2, t2);
 482     _mm_storeu_pd(ptrA+4, t3);
 483     _mm_storeu_pd(ptrA+6, t4);
 484     _mm_store_sd(ptrA+8, t5);
 485 }
 486 #endif
 487
 488
 489 #if defined (_MSC_VER) && defined(_M_IX86)
 490 /* Macro work-around since 32-bit MSVC cannot handle >3 xmm/ymm parameters */
 491 #define gmx_mm_decrement_4rvec_1ptr_swizzle_pd(ptrA, _x1, _y1, _z1, _x2, _y2, _z2, _x3, _y3, _z3, _x4, _y4, _z4) \
 492     { \
 493         __m128d _t1, _t2, _t3, _t4, _t5, _t6; \
 494         _t1          = _mm_loadu_pd(ptrA); \
 495         _t2          = _mm_loadu_pd(ptrA+2); \
 496         _t3          = _mm_loadu_pd(ptrA+4); \
 497         _t4          = _mm_loadu_pd(ptrA+6); \
 498         _t5          = _mm_loadu_pd(ptrA+8); \
 499         _t6          = _mm_loadu_pd(ptrA+10); \
 500         _x1          = _mm_unpacklo_pd(_x1, _y1); \
 501         _z1          = _mm_unpacklo_pd(_z1, _x2); \
 502         _y2          = _mm_unpacklo_pd(_y2, _z2); \
 503         _x3          = _mm_unpacklo_pd(_x3, _y3); \
 504         _z3          = _mm_unpacklo_pd(_z3, _x4); \
 505         _y4          = _mm_unpacklo_pd(_y4, _z4); \
 506         _mm_storeu_pd(ptrA,    _mm_sub_pd( _t1, _x1 )); \
 507         _mm_storeu_pd(ptrA+2,  _mm_sub_pd( _t2, _z1 )); \
 508         _mm_storeu_pd(ptrA+4,  _mm_sub_pd( _t3, _y2 )); \
 509         _mm_storeu_pd(ptrA+6,  _mm_sub_pd( _t4, _x3 )); \
 510         _mm_storeu_pd(ptrA+8,  _mm_sub_pd( _t5, _z3 )); \
 511         _mm_storeu_pd(ptrA+10, _mm_sub_pd( _t6, _y4 )); \
 512     }
 513 #else
 514 /* Real function for sane compilers */
 515 static gmx_inline void
 516 gmx_mm_decrement_4rvec_1ptr_swizzle_pd(double * gmx_restrict ptrA,
 517                                        __m128d x1, __m128d y1, __m128d z1,
 518                                        __m128d x2, __m128d y2, __m128d z2,
 519                                        __m128d x3, __m128d y3, __m128d z3,
 520                                        __m128d x4, __m128d y4, __m128d z4)
 521 {
 522     __m128d t1, t2, t3, t4, t5, t6;
 523
 524     t1          = _mm_loadu_pd(ptrA);
 525     t2          = _mm_loadu_pd(ptrA+2);
 526     t3          = _mm_loadu_pd(ptrA+4);
 527     t4          = _mm_loadu_pd(ptrA+6);
 528     t5          = _mm_loadu_pd(ptrA+8);
 529     t6          = _mm_loadu_pd(ptrA+10);
 530
 531     x1          = _mm_unpacklo_pd(x1, y1);
 532     z1          = _mm_unpacklo_pd(z1, x2);
 533     y2          = _mm_unpacklo_pd(y2, z2);
 534     x3          = _mm_unpacklo_pd(x3, y3);
 535     z3          = _mm_unpacklo_pd(z3, x4);
 536     y4          = _mm_unpacklo_pd(y4, z4);
 537
 538     _mm_storeu_pd(ptrA,    _mm_sub_pd( t1, x1 ));
 539     _mm_storeu_pd(ptrA+2,  _mm_sub_pd( t2, z1 ));
 540     _mm_storeu_pd(ptrA+4,  _mm_sub_pd( t3, y2 ));
 541     _mm_storeu_pd(ptrA+6,  _mm_sub_pd( t4, x3 ));
 542     _mm_storeu_pd(ptrA+8,  _mm_sub_pd( t5, z3 ));
 543     _mm_storeu_pd(ptrA+10, _mm_sub_pd( t6, y4 ));
 544 }
 545 #endif
 546
 547
 548 static gmx_inline void
 549 gmx_mm_decrement_1rvec_2ptr_swizzle_pd(double * gmx_restrict ptrA, double * gmx_restrict ptrB,
 550                                        __m128d x1, __m128d y1, __m128d z1)
 551 {
 552     __m128d t1, t2, t3, t4, t5, t6, t7;
 553
 554     t1          = _mm_loadu_pd(ptrA);
 555     t2          = _mm_load_sd(ptrA+2);
 556     t3          = _mm_loadu_pd(ptrB);
 557     t4          = _mm_load_sd(ptrB+2);
 558
 559     t5          = _mm_unpacklo_pd(x1, y1);
 560     t6          = _mm_unpackhi_pd(x1, y1);
 561     t7          = _mm_unpackhi_pd(z1, z1);
 562
 563     t1          = _mm_sub_pd(t1, t5);
 564     t2          = _mm_sub_sd(t2, z1);
 565
 566     t3          = _mm_sub_pd(t3, t6);
 567     t4          = _mm_sub_sd(t4, t7);
 568
 569     _mm_storeu_pd(ptrA, t1);
 570     _mm_store_sd(ptrA+2, t2);
 571     _mm_storeu_pd(ptrB, t3);
 572     _mm_store_sd(ptrB+2, t4);
 573 }
 574
 575 #if defined (_MSC_VER) && defined(_M_IX86)
 576 /* Macro work-around since 32-bit MSVC cannot handle >3 xmm/ymm parameters */
 577 #define gmx_mm_decrement_3rvec_2ptr_swizzle_pd(ptrA, ptrB, _x1, _y1, _z1, _x2, _y2, _z2, _x3, _y3, _z3) \
 578     { \
 579         __m128d _t1, _t2, _t3, _t4, _t5, _t6, _t7, _t8, _t9, _t10; \
 580         __m128d _tA, _tB, _tC, _tD, _tE, _tF, _tG, _tH, _tI; \
 581         _t1          = _mm_loadu_pd(ptrA); \
 582         _t2          = _mm_loadu_pd(ptrA+2); \
 583         _t3          = _mm_loadu_pd(ptrA+4); \
 584         _t4          = _mm_loadu_pd(ptrA+6); \
 585         _t5          = _mm_load_sd(ptrA+8); \
 586         _t6          = _mm_loadu_pd(ptrB); \
 587         _t7          = _mm_loadu_pd(ptrB+2); \
 588         _t8          = _mm_loadu_pd(ptrB+4); \
 589         _t9          = _mm_loadu_pd(ptrB+6); \
 590         _t10         = _mm_load_sd(ptrB+8); \
 591         _tA          = _mm_unpacklo_pd(_x1, _y1); \
 592         _tB          = _mm_unpackhi_pd(_x1, _y1); \
 593         _tC          = _mm_unpacklo_pd(_z1, _x2); \
 594         _tD          = _mm_unpackhi_pd(_z1, _x2); \
 595         _tE          = _mm_unpacklo_pd(_y2, _z2); \
 596         _tF          = _mm_unpackhi_pd(_y2, _z2); \
 597         _tG          = _mm_unpacklo_pd(_x3, _y3); \
 598         _tH          = _mm_unpackhi_pd(_x3, _y3); \
 599         _tI          = _mm_unpackhi_pd(_z3, _z3); \
 600         _t1          = _mm_sub_pd(_t1, _tA); \
 601         _t2          = _mm_sub_pd(_t2, _tC); \
 602         _t3          = _mm_sub_pd(_t3, _tE); \
 603         _t4          = _mm_sub_pd(_t4, _tG); \
 604         _t5          = _mm_sub_sd(_t5, _z3); \
 605         _t6          = _mm_sub_pd(_t6, _tB); \
 606         _t7          = _mm_sub_pd(_t7, _tD); \
 607         _t8          = _mm_sub_pd(_t8, _tF); \
 608         _t9          = _mm_sub_pd(_t9, _tH); \
 609         _t10         = _mm_sub_sd(_t10, _tI); \
 610         _mm_storeu_pd(ptrA, _t1); \
 611         _mm_storeu_pd(ptrA+2, _t2); \
 612         _mm_storeu_pd(ptrA+4, _t3); \
 613         _mm_storeu_pd(ptrA+6, _t4); \
 614         _mm_store_sd(ptrA+8, _t5); \
 615         _mm_storeu_pd(ptrB, _t6); \
 616         _mm_storeu_pd(ptrB+2, _t7); \
 617         _mm_storeu_pd(ptrB+4, _t8); \
 618         _mm_storeu_pd(ptrB+6, _t9); \
 619         _mm_store_sd(ptrB+8, _t10); \
 620     }
 621 #else
 622 /* Real function for sane compilers */
 623 static gmx_inline void
 624 gmx_mm_decrement_3rvec_2ptr_swizzle_pd(double * gmx_restrict ptrA, double * gmx_restrict ptrB,
 625                                        __m128d x1, __m128d y1, __m128d z1,
 626                                        __m128d x2, __m128d y2, __m128d z2,
 627                                        __m128d x3, __m128d y3, __m128d z3)
 628 {
 629     __m128d t1, t2, t3, t4, t5, t6, t7, t8, t9, t10;
 630     __m128d tA, tB, tC, tD, tE, tF, tG, tH, tI;
 631
 632     t1          = _mm_loadu_pd(ptrA);
 633     t2          = _mm_loadu_pd(ptrA+2);
 634     t3          = _mm_loadu_pd(ptrA+4);
 635     t4          = _mm_loadu_pd(ptrA+6);
 636     t5          = _mm_load_sd(ptrA+8);
 637     t6          = _mm_loadu_pd(ptrB);
 638     t7          = _mm_loadu_pd(ptrB+2);
 639     t8          = _mm_loadu_pd(ptrB+4);
 640     t9          = _mm_loadu_pd(ptrB+6);
 641     t10         = _mm_load_sd(ptrB+8);
 642
 643     tA          = _mm_unpacklo_pd(x1, y1);
 644     tB          = _mm_unpackhi_pd(x1, y1);
 645     tC          = _mm_unpacklo_pd(z1, x2);
 646     tD          = _mm_unpackhi_pd(z1, x2);
 647     tE          = _mm_unpacklo_pd(y2, z2);
 648     tF          = _mm_unpackhi_pd(y2, z2);
 649     tG          = _mm_unpacklo_pd(x3, y3);
 650     tH          = _mm_unpackhi_pd(x3, y3);
 651     tI          = _mm_unpackhi_pd(z3, z3);
 652
 653     t1          = _mm_sub_pd(t1, tA);
 654     t2          = _mm_sub_pd(t2, tC);
 655     t3          = _mm_sub_pd(t3, tE);
 656     t4          = _mm_sub_pd(t4, tG);
 657     t5          = _mm_sub_sd(t5, z3);
 658
 659     t6          = _mm_sub_pd(t6, tB);
 660     t7          = _mm_sub_pd(t7, tD);
 661     t8          = _mm_sub_pd(t8, tF);
 662     t9          = _mm_sub_pd(t9, tH);
 663     t10         = _mm_sub_sd(t10, tI);
 664
 665     _mm_storeu_pd(ptrA, t1);
 666     _mm_storeu_pd(ptrA+2, t2);
 667     _mm_storeu_pd(ptrA+4, t3);
 668     _mm_storeu_pd(ptrA+6, t4);
 669     _mm_store_sd(ptrA+8, t5);
 670     _mm_storeu_pd(ptrB, t6);
 671     _mm_storeu_pd(ptrB+2, t7);
 672     _mm_storeu_pd(ptrB+4, t8);
 673     _mm_storeu_pd(ptrB+6, t9);
 674     _mm_store_sd(ptrB+8, t10);
 675 }
 676 #endif
 677
 678
 679 #if defined (_MSC_VER) && defined(_M_IX86)
 680 /* Macro work-around since 32-bit MSVC cannot handle >3 xmm/ymm parameters */
 681 #define gmx_mm_decrement_4rvec_2ptr_swizzle_pd(ptrA, ptrB, _x1, _y1, _z1, _x2, _y2, _z2, _x3, _y3, _z3, _x4, _y4, _z4) \
 682     { \
 683         __m128d _t1, _t2, _t3, _t4, _t5, _t6, _t7, _t8, _t9, _t10, _t11, _t12; \
 684         __m128d _tA, _tB, _tC, _tD, _tE, _tF, _tG, _tH, _tI, _tJ, _tK, _tL; \
 685         _t1          = _mm_loadu_pd(ptrA); \
 686         _t2          = _mm_loadu_pd(ptrA+2); \
 687         _t3          = _mm_loadu_pd(ptrA+4); \
 688         _t4          = _mm_loadu_pd(ptrA+6); \
 689         _t5          = _mm_loadu_pd(ptrA+8); \
 690         _t6          = _mm_loadu_pd(ptrA+10); \
 691         _t7          = _mm_loadu_pd(ptrB); \
 692         _t8          = _mm_loadu_pd(ptrB+2); \
 693         _t9          = _mm_loadu_pd(ptrB+4); \
 694         _t10         = _mm_loadu_pd(ptrB+6); \
 695         _t11         = _mm_loadu_pd(ptrB+8); \
 696         _t12         = _mm_loadu_pd(ptrB+10); \
 697         _tA          = _mm_unpacklo_pd(_x1, _y1); \
 698         _tB          = _mm_unpackhi_pd(_x1, _y1); \
 699         _tC          = _mm_unpacklo_pd(_z1, _x2); \
 700         _tD          = _mm_unpackhi_pd(_z1, _x2); \
 701         _tE          = _mm_unpacklo_pd(_y2, _z2); \
 702         _tF          = _mm_unpackhi_pd(_y2, _z2); \
 703         _tG          = _mm_unpacklo_pd(_x3, _y3); \
 704         _tH          = _mm_unpackhi_pd(_x3, _y3); \
 705         _tI          = _mm_unpacklo_pd(_z3, _x4); \
 706         _tJ          = _mm_unpackhi_pd(_z3, _x4); \
 707         _tK          = _mm_unpacklo_pd(_y4, _z4); \
 708         _tL          = _mm_unpackhi_pd(_y4, _z4); \
 709         _t1          = _mm_sub_pd(_t1, _tA); \
 710         _t2          = _mm_sub_pd(_t2, _tC); \
 711         _t3          = _mm_sub_pd(_t3, _tE); \
 712         _t4          = _mm_sub_pd(_t4, _tG); \
 713         _t5          = _mm_sub_pd(_t5, _tI); \
 714         _t6          = _mm_sub_pd(_t6, _tK); \
 715         _t7          = _mm_sub_pd(_t7, _tB); \
 716         _t8          = _mm_sub_pd(_t8, _tD); \
 717         _t9          = _mm_sub_pd(_t9, _tF); \
 718         _t10         = _mm_sub_pd(_t10, _tH); \
 719         _t11         = _mm_sub_pd(_t11, _tJ); \
 720         _t12         = _mm_sub_pd(_t12, _tL); \
 721         _mm_storeu_pd(ptrA,  _t1); \
 722         _mm_storeu_pd(ptrA+2, _t2); \
 723         _mm_storeu_pd(ptrA+4, _t3); \
 724         _mm_storeu_pd(ptrA+6, _t4); \
 725         _mm_storeu_pd(ptrA+8, _t5); \
 726         _mm_storeu_pd(ptrA+10, _t6); \
 727         _mm_storeu_pd(ptrB,  _t7); \
 728         _mm_storeu_pd(ptrB+2, _t8); \
 729         _mm_storeu_pd(ptrB+4, _t9); \
 730         _mm_storeu_pd(ptrB+6, _t10); \
 731         _mm_storeu_pd(ptrB+8, _t11); \
 732         _mm_storeu_pd(ptrB+10, _t12); \
 733     }
 734 #else
 735 /* Real function for sane compilers */
 736 static gmx_inline void
 737 gmx_mm_decrement_4rvec_2ptr_swizzle_pd(double * gmx_restrict ptrA, double * gmx_restrict ptrB,
 738                                        __m128d x1, __m128d y1, __m128d z1,
 739                                        __m128d x2, __m128d y2, __m128d z2,
 740                                        __m128d x3, __m128d y3, __m128d z3,
 741                                        __m128d x4, __m128d y4, __m128d z4)
 742 {
 743     __m128d t1, t2, t3, t4, t5, t6, t7, t8, t9, t10, t11, t12;
 744     __m128d tA, tB, tC, tD, tE, tF, tG, tH, tI, tJ, tK, tL;
 745
 746     t1          = _mm_loadu_pd(ptrA);
 747     t2          = _mm_loadu_pd(ptrA+2);
 748     t3          = _mm_loadu_pd(ptrA+4);
 749     t4          = _mm_loadu_pd(ptrA+6);
 750     t5          = _mm_loadu_pd(ptrA+8);
 751     t6          = _mm_loadu_pd(ptrA+10);
 752     t7          = _mm_loadu_pd(ptrB);
 753     t8          = _mm_loadu_pd(ptrB+2);
 754     t9          = _mm_loadu_pd(ptrB+4);
 755     t10         = _mm_loadu_pd(ptrB+6);
 756     t11         = _mm_loadu_pd(ptrB+8);
 757     t12         = _mm_loadu_pd(ptrB+10);
 758
 759     tA          = _mm_unpacklo_pd(x1, y1);
 760     tB          = _mm_unpackhi_pd(x1, y1);
 761     tC          = _mm_unpacklo_pd(z1, x2);
 762     tD          = _mm_unpackhi_pd(z1, x2);
 763     tE          = _mm_unpacklo_pd(y2, z2);
 764     tF          = _mm_unpackhi_pd(y2, z2);
 765     tG          = _mm_unpacklo_pd(x3, y3);
 766     tH          = _mm_unpackhi_pd(x3, y3);
 767     tI          = _mm_unpacklo_pd(z3, x4);
 768     tJ          = _mm_unpackhi_pd(z3, x4);
 769     tK          = _mm_unpacklo_pd(y4, z4);
 770     tL          = _mm_unpackhi_pd(y4, z4);
 771
 772     t1          = _mm_sub_pd(t1, tA);
 773     t2          = _mm_sub_pd(t2, tC);
 774     t3          = _mm_sub_pd(t3, tE);
 775     t4          = _mm_sub_pd(t4, tG);
 776     t5          = _mm_sub_pd(t5, tI);
 777     t6          = _mm_sub_pd(t6, tK);
 778
 779     t7          = _mm_sub_pd(t7, tB);
 780     t8          = _mm_sub_pd(t8, tD);
 781     t9          = _mm_sub_pd(t9, tF);
 782     t10         = _mm_sub_pd(t10, tH);
 783     t11         = _mm_sub_pd(t11, tJ);
 784     t12         = _mm_sub_pd(t12, tL);
 785
 786     _mm_storeu_pd(ptrA,  t1);
 787     _mm_storeu_pd(ptrA+2, t2);
 788     _mm_storeu_pd(ptrA+4, t3);
 789     _mm_storeu_pd(ptrA+6, t4);
 790     _mm_storeu_pd(ptrA+8, t5);
 791     _mm_storeu_pd(ptrA+10, t6);
 792     _mm_storeu_pd(ptrB,  t7);
 793     _mm_storeu_pd(ptrB+2, t8);
 794     _mm_storeu_pd(ptrB+4, t9);
 795     _mm_storeu_pd(ptrB+6, t10);
 796     _mm_storeu_pd(ptrB+8, t11);
 797     _mm_storeu_pd(ptrB+10, t12);
 798 }
 799 #endif
 800
 801
 802
 803
 804 static gmx_inline void
 805 gmx_mm_update_iforce_1atom_swizzle_pd(__m128d fix1, __m128d fiy1, __m128d fiz1,
 806                                       double * gmx_restrict fptr,
 807                                       double * gmx_restrict fshiftptr)
 808 {
 809     fix1 = _mm_hadd_pd(fix1, fiy1);
 810     fiz1 = _mm_hadd_pd(fiz1, fiz1);
 811
 812     _mm_storeu_pd( fptr, _mm_add_pd( _mm_loadu_pd(fptr), fix1 ));
 813     _mm_store_sd( fptr+2, _mm_add_sd( _mm_load_sd(fptr+2), fiz1 ));
 814
 815     _mm_storeu_pd( fshiftptr, _mm_add_pd( _mm_loadu_pd(fshiftptr), fix1 ));
 816     _mm_store_sd( fshiftptr+2, _mm_add_sd( _mm_load_sd(fshiftptr+2), fiz1 ));
 817 }
 818
 819
 820 #if defined (_MSC_VER) && defined(_M_IX86)
 821 /* Macro work-around since 32-bit MSVC cannot handle >3 xmm/ymm parameters */
 822 #define gmx_mm_update_iforce_3atom_swizzle_pd(fix1, fiy1, fiz1, fix2, fiy2, fiz2, fix3, fiy3, fiz3, \
 823                                               fptr, fshiftptr) \
 824     { \
 825         __m128d _t1, _t2; \
 826         fix1 = _mm_hadd_pd(fix1, fiy1); \
 827         fiz1 = _mm_hadd_pd(fiz1, fix2); \
 828         fiy2 = _mm_hadd_pd(fiy2, fiz2); \
 829         fix3 = _mm_hadd_pd(fix3, fiy3); \
 830         fiz3 = _mm_hadd_pd(fiz3, fiz3); \
 831         _mm_storeu_pd( fptr, _mm_add_pd( _mm_loadu_pd(fptr), fix1 )); \
 832         _mm_storeu_pd( fptr+2, _mm_add_pd( _mm_loadu_pd(fptr+2), fiz1 )); \
 833         _mm_storeu_pd( fptr+4, _mm_add_pd( _mm_loadu_pd(fptr+4), fiy2 )); \
 834         _mm_storeu_pd( fptr+6, _mm_add_pd( _mm_loadu_pd(fptr+6), fix3 )); \
 835         _mm_store_sd( fptr+8, _mm_add_sd( _mm_load_sd(fptr+8), fiz3 )); \
 836         fix1  = _mm_add_pd(fix1, fix3); \
 837         _t1   = _mm_shuffle_pd(fiz1, fiy2, _MM_SHUFFLE2(0, 1)); \
 838         fix1  = _mm_add_pd(fix1, _t1); \
 839         _t2   = _mm_shuffle_pd(fiy2, fiy2, _MM_SHUFFLE2(1, 1)); \
 840         fiz1  = _mm_add_sd(fiz1, fiz3); \
 841         fiz1  = _mm_add_sd(fiz1, _t2); \
 842         _mm_storeu_pd( fshiftptr, _mm_add_pd( _mm_loadu_pd(fshiftptr), fix1 )); \
 843         _mm_store_sd( fshiftptr+2, _mm_add_sd( _mm_load_sd(fshiftptr+2), fiz1 )); \
 844     }
 845 #else
 846 /* Real function for sane compilers */
 847 static gmx_inline void
 848 gmx_mm_update_iforce_3atom_swizzle_pd(__m128d fix1, __m128d fiy1, __m128d fiz1,
 849                                       __m128d fix2, __m128d fiy2, __m128d fiz2,
 850                                       __m128d fix3, __m128d fiy3, __m128d fiz3,
 851                                       double * gmx_restrict fptr,
 852                                       double * gmx_restrict fshiftptr)
 853 {
 854     __m128d t1, t2;
 855
 856     fix1 = _mm_hadd_pd(fix1, fiy1);
 857     fiz1 = _mm_hadd_pd(fiz1, fix2);
 858     fiy2 = _mm_hadd_pd(fiy2, fiz2);
 859     fix3 = _mm_hadd_pd(fix3, fiy3);
 860     fiz3 = _mm_hadd_pd(fiz3, fiz3);
 861
 862     _mm_storeu_pd( fptr, _mm_add_pd( _mm_loadu_pd(fptr), fix1 ));
 863     _mm_storeu_pd( fptr+2, _mm_add_pd( _mm_loadu_pd(fptr+2), fiz1 ));
 864     _mm_storeu_pd( fptr+4, _mm_add_pd( _mm_loadu_pd(fptr+4), fiy2 ));
 865     _mm_storeu_pd( fptr+6, _mm_add_pd( _mm_loadu_pd(fptr+6), fix3 ));
 866     _mm_store_sd( fptr+8, _mm_add_sd( _mm_load_sd(fptr+8), fiz3 ));
 867
 868     fix1 = _mm_add_pd(fix1, fix3);
 869     t1   = _mm_shuffle_pd(fiz1, fiy2, _MM_SHUFFLE2(0, 1));
 870     fix1 = _mm_add_pd(fix1, t1); /* x and y sums */
 871
 872     t2   = _mm_shuffle_pd(fiy2, fiy2, _MM_SHUFFLE2(1, 1));
 873     fiz1 = _mm_add_sd(fiz1, fiz3);
 874     fiz1 = _mm_add_sd(fiz1, t2); /* z sum */
 875
 876     _mm_storeu_pd( fshiftptr, _mm_add_pd( _mm_loadu_pd(fshiftptr), fix1 ));
 877     _mm_store_sd( fshiftptr+2, _mm_add_sd( _mm_load_sd(fshiftptr+2), fiz1 ));
 878 }
 879 #endif
 880
 881 #if defined (_MSC_VER) && defined(_M_IX86)
 882 /* Macro work-around since 32-bit MSVC cannot handle >3 xmm/ymm parameters */
 883 #define gmx_mm_update_iforce_4atom_swizzle_pd(fix1, fiy1, fiz1, fix2, fiy2, fiz2, fix3, fiy3, fiz3, fix4, fiy4, fiz4, \
 884                                               fptr, fshiftptr) \
 885     { \
 886         __m128d _t1, _t2; \
 887         fix1 = _mm_hadd_pd(fix1, fiy1); \
 888         fiz1 = _mm_hadd_pd(fiz1, fix2); \
 889         fiy2 = _mm_hadd_pd(fiy2, fiz2); \
 890         fix3 = _mm_hadd_pd(fix3, fiy3); \
 891         fiz3 = _mm_hadd_pd(fiz3, fix4); \
 892         fiy4 = _mm_hadd_pd(fiy4, fiz4); \
 893         _mm_storeu_pd( fptr, _mm_add_pd( _mm_loadu_pd(fptr),       fix1 )); \
 894         _mm_storeu_pd( fptr+2, _mm_add_pd( _mm_loadu_pd(fptr+2),   fiz1 )); \
 895         _mm_storeu_pd( fptr+4, _mm_add_pd( _mm_loadu_pd(fptr+4),   fiy2 )); \
 896         _mm_storeu_pd( fptr+6, _mm_add_pd( _mm_loadu_pd(fptr+6),   fix3 )); \
 897         _mm_storeu_pd( fptr+8, _mm_add_pd( _mm_loadu_pd(fptr+8),   fiz3 )); \
 898         _mm_storeu_pd( fptr+10, _mm_add_pd( _mm_loadu_pd(fptr+10), fiy4 )); \
 899         _t1  = _mm_shuffle_pd(fiz1, fiy2, _MM_SHUFFLE2(0, 1)); \
 900         fix1 = _mm_add_pd(fix1, _t1); \
 901         _t2  = _mm_shuffle_pd(fiz3, fiy4, _MM_SHUFFLE2(0, 1)); \
 902         fix3 = _mm_add_pd(fix3, _t2); \
 903         fix1 = _mm_add_pd(fix1, fix3); \
 904         fiz1 = _mm_add_sd(fiz1, _mm_unpackhi_pd(fiy2, fiy2)); \
 905         fiz3 = _mm_add_sd(fiz3, _mm_unpackhi_pd(fiy4, fiy4)); \
 906         fiz1 = _mm_add_sd(fiz1, fiz3); \
 907         _mm_storeu_pd( fshiftptr, _mm_add_pd( _mm_loadu_pd(fshiftptr), fix1 )); \
 908         _mm_store_sd( fshiftptr+2, _mm_add_sd( _mm_load_sd(fshiftptr+2), fiz1 )); \
 909     }
 910 #else
 911 /* Real function for sane compilers */
 912 static gmx_inline void
 913 gmx_mm_update_iforce_4atom_swizzle_pd(__m128d fix1, __m128d fiy1, __m128d fiz1,
 914                                       __m128d fix2, __m128d fiy2, __m128d fiz2,
 915                                       __m128d fix3, __m128d fiy3, __m128d fiz3,
 916                                       __m128d fix4, __m128d fiy4, __m128d fiz4,
 917                                       double * gmx_restrict fptr,
 918                                       double * gmx_restrict fshiftptr)
 919 {
 920     __m128d t1, t2;
 921
 922     fix1 = _mm_hadd_pd(fix1, fiy1);
 923     fiz1 = _mm_hadd_pd(fiz1, fix2);
 924     fiy2 = _mm_hadd_pd(fiy2, fiz2);
 925     fix3 = _mm_hadd_pd(fix3, fiy3);
 926     fiz3 = _mm_hadd_pd(fiz3, fix4);
 927     fiy4 = _mm_hadd_pd(fiy4, fiz4);
 928
 929     _mm_storeu_pd( fptr, _mm_add_pd( _mm_loadu_pd(fptr),       fix1 ));
 930     _mm_storeu_pd( fptr+2, _mm_add_pd( _mm_loadu_pd(fptr+2),   fiz1 ));
 931     _mm_storeu_pd( fptr+4, _mm_add_pd( _mm_loadu_pd(fptr+4),   fiy2 ));
 932     _mm_storeu_pd( fptr+6, _mm_add_pd( _mm_loadu_pd(fptr+6),   fix3 ));
 933     _mm_storeu_pd( fptr+8, _mm_add_pd( _mm_loadu_pd(fptr+8),   fiz3 ));
 934     _mm_storeu_pd( fptr+10, _mm_add_pd( _mm_loadu_pd(fptr+10), fiy4 ));
 935
 936     t1   = _mm_shuffle_pd(fiz1, fiy2, _MM_SHUFFLE2(0, 1));
 937     fix1 = _mm_add_pd(fix1, t1);
 938     t2   = _mm_shuffle_pd(fiz3, fiy4, _MM_SHUFFLE2(0, 1));
 939     fix3 = _mm_add_pd(fix3, t2);
 940     fix1 = _mm_add_pd(fix1, fix3); /* x and y sums */
 941
 942     fiz1 = _mm_add_sd(fiz1, _mm_unpackhi_pd(fiy2, fiy2));
 943     fiz3 = _mm_add_sd(fiz3, _mm_unpackhi_pd(fiy4, fiy4));
 944     fiz1 = _mm_add_sd(fiz1, fiz3); /* z sum */
 945
 946     _mm_storeu_pd( fshiftptr, _mm_add_pd( _mm_loadu_pd(fshiftptr), fix1 ));
 947     _mm_store_sd( fshiftptr+2, _mm_add_sd( _mm_load_sd(fshiftptr+2), fiz1 ));
 948 }
 949 #endif
 950
 951 static gmx_inline void
 952 gmx_mm_update_1pot_pd(__m128d pot1, double * gmx_restrict ptrA)
 953 {
 954     pot1 = _mm_hadd_pd(pot1, pot1);
 955     _mm_store_sd(ptrA, _mm_add_sd(pot1, _mm_load_sd(ptrA)));
 956 }
 957
 958 static gmx_inline void
 959 gmx_mm_update_2pot_pd(__m128d pot1, double * gmx_restrict ptrA,
 960                       __m128d pot2, double * gmx_restrict ptrB)
 961 {
 962     pot1 = _mm_hadd_pd(pot1, pot2);
 963     pot2 = _mm_unpackhi_pd(pot1, pot1);
 964
 965     _mm_store_sd(ptrA, _mm_add_sd(pot1, _mm_load_sd(ptrA)));
 966     _mm_store_sd(ptrB, _mm_add_sd(pot2, _mm_load_sd(ptrB)));
 967 }
 968
 969
 970 #endif /* _kernelutil_x86_sse4_1_double_h_ */