src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/kernelutil_x86_sse2_single.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_sse2_single_h_
  36 #define _kernelutil_x86_sse2_single_h_
  37
  38 /* We require SSE2 now! */
  39
  40 #include <math.h>
  41
  42 #include "gromacs/simd/general_x86_sse2.h"
  43
  44
  45 /* Normal sum of four xmm registers */
  46 #define gmx_mm_sum4_ps(t0, t1, t2, t3)  _mm_add_ps(_mm_add_ps(t0, t1), _mm_add_ps(t2, t3))
  47
  48 static gmx_inline __m128
  49 gmx_mm_calc_rsq_ps(__m128 dx, __m128 dy, __m128 dz)
  50 {
  51     return _mm_add_ps( _mm_add_ps( _mm_mul_ps(dx, dx), _mm_mul_ps(dy, dy) ), _mm_mul_ps(dz, dz) );
  52 }
  53
  54 static int
  55 gmx_mm_any_lt(__m128 a, __m128 b)
  56 {
  57     return _mm_movemask_ps(_mm_cmplt_ps(a, b));
  58 }
  59
  60 /* Load a single value from 1-4 places, merge into xmm register */
  61
  62 static __m128
  63 gmx_mm_load_4real_swizzle_ps(const float * gmx_restrict ptrA,
  64                              const float * gmx_restrict ptrB,
  65                              const float * gmx_restrict ptrC,
  66                              const float * gmx_restrict ptrD)
  67 {
  68     __m128 t1, t2;
  69
  70     t1 = _mm_unpacklo_ps(_mm_load_ss(ptrA), _mm_load_ss(ptrC));
  71     t2 = _mm_unpacklo_ps(_mm_load_ss(ptrB), _mm_load_ss(ptrD));
  72     return _mm_unpacklo_ps(t1, t2);
  73 }
  74
  75 static void
  76 gmx_mm_store_4real_swizzle_ps(float * gmx_restrict ptrA,
  77                               float * gmx_restrict ptrB,
  78                               float * gmx_restrict ptrC,
  79                               float * gmx_restrict ptrD,
  80                               __m128               xmm1)
  81 {
  82     __m128 t2, t3, t4;
  83
  84     t3       = _mm_movehl_ps(_mm_setzero_ps(), xmm1);
  85     t2       = _mm_shuffle_ps(xmm1, xmm1, _MM_SHUFFLE(1, 1, 1, 1));
  86     t4       = _mm_shuffle_ps(t3, t3, _MM_SHUFFLE(1, 1, 1, 1));
  87     _mm_store_ss(ptrA, xmm1);
  88     _mm_store_ss(ptrB, t2);
  89     _mm_store_ss(ptrC, t3);
  90     _mm_store_ss(ptrD, t4);
  91 }
  92
  93 /* Similar to store, but increments value in memory */
  94 static void
  95 gmx_mm_increment_4real_swizzle_ps(float * gmx_restrict ptrA,
  96                                   float * gmx_restrict ptrB,
  97                                   float * gmx_restrict ptrC,
  98                                   float * gmx_restrict ptrD, __m128 xmm1)
  99 {
 100     __m128 tmp;
 101
 102     tmp = gmx_mm_load_4real_swizzle_ps(ptrA, ptrB, ptrC, ptrD);
 103     tmp = _mm_add_ps(tmp, xmm1);
 104     gmx_mm_store_4real_swizzle_ps(ptrA, ptrB, ptrC, ptrD, tmp);
 105 }
 106
 107
 108 static void
 109 gmx_mm_load_4pair_swizzle_ps(const float * gmx_restrict p1,
 110                              const float * gmx_restrict p2,
 111                              const float * gmx_restrict p3,
 112                              const float * gmx_restrict p4,
 113                              __m128 * gmx_restrict      c6,
 114                              __m128 * gmx_restrict      c12)
 115 {
 116     __m128 t1, t2, t3, t4;
 117
 118     t1   = _mm_loadl_pi(_mm_setzero_ps(), (__m64 *)p1);   /* - - c12a  c6a */
 119     t2   = _mm_loadl_pi(_mm_setzero_ps(), (__m64 *)p2);   /* - - c12b  c6b */
 120     t3   = _mm_loadl_pi(_mm_setzero_ps(), (__m64 *)p3);   /* - - c12c  c6c */
 121     t4   = _mm_loadl_pi(_mm_setzero_ps(), (__m64 *)p4);   /* - - c12d  c6d */
 122     t1   = _mm_unpacklo_ps(t1, t2);
 123     t2   = _mm_unpacklo_ps(t3, t4);
 124     *c6  = _mm_movelh_ps(t1, t2);
 125     *c12 = _mm_movehl_ps(t2, t1);
 126 }
 127
 128 /* Routines to load 1-4 rvec from 4 places.
 129  * We mainly use these to load coordinates. The extra routines
 130  * are very efficient for the water-water loops, since we e.g.
 131  * know that a TIP4p water has 4 atoms, so we should load 12 floats+shuffle.
 132  */
 133
 134
 135 static gmx_inline void
 136 gmx_mm_load_shift_and_1rvec_broadcast_ps(const float * gmx_restrict xyz_shift,
 137                                          const float * gmx_restrict xyz,
 138                                          __m128 * gmx_restrict      x1,
 139                                          __m128 * gmx_restrict      y1,
 140                                          __m128 * gmx_restrict      z1)
 141 {
 142     __m128 t1, t2, t3, t4;
 143
 144     t1   = _mm_loadl_pi(_mm_setzero_ps(), (__m64 *)xyz_shift);
 145     t2   = _mm_loadl_pi(_mm_setzero_ps(), (__m64 *)xyz);
 146     t3   = _mm_load_ss(xyz_shift+2);
 147     t4   = _mm_load_ss(xyz+2);
 148     t1   = _mm_add_ps(t1, t2);
 149     t3   = _mm_add_ss(t3, t4);
 150
 151     *x1  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(0, 0, 0, 0));
 152     *y1  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(1, 1, 1, 1));
 153     *z1  = _mm_shuffle_ps(t3, t3, _MM_SHUFFLE(0, 0, 0, 0));
 154 }
 155
 156
 157 static gmx_inline void
 158 gmx_mm_load_shift_and_3rvec_broadcast_ps(const float * gmx_restrict xyz_shift,
 159                                          const float * gmx_restrict xyz,
 160                                          __m128 * gmx_restrict x1, __m128 * gmx_restrict y1, __m128 * gmx_restrict z1,
 161                                          __m128 * gmx_restrict x2, __m128 * gmx_restrict y2, __m128 * gmx_restrict z2,
 162                                          __m128 * gmx_restrict x3, __m128 * gmx_restrict y3, __m128 * gmx_restrict z3)
 163 {
 164     __m128 tA, tB;
 165     __m128 t1, t2, t3, t4, t5, t6;
 166
 167     tA   = _mm_loadl_pi(_mm_setzero_ps(), (__m64 *)xyz_shift);
 168     tB   = _mm_load_ss(xyz_shift+2);
 169
 170     t1   = _mm_loadu_ps(xyz);
 171     t2   = _mm_loadu_ps(xyz+4);
 172     t3   = _mm_load_ss(xyz+8);
 173
 174     tA   = _mm_movelh_ps(tA, tB);
 175     t4   = _mm_shuffle_ps(tA, tA, _MM_SHUFFLE(0, 2, 1, 0));
 176     t5   = _mm_shuffle_ps(tA, tA, _MM_SHUFFLE(1, 0, 2, 1));
 177     t6   = _mm_shuffle_ps(tA, tA, _MM_SHUFFLE(2, 1, 0, 2));
 178
 179     t1   = _mm_add_ps(t1, t4);
 180     t2   = _mm_add_ps(t2, t5);
 181     t3   = _mm_add_ss(t3, t6);
 182
 183     *x1  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(0, 0, 0, 0));
 184     *y1  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(1, 1, 1, 1));
 185     *z1  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(2, 2, 2, 2));
 186     *x2  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(3, 3, 3, 3));
 187     *y2  = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(0, 0, 0, 0));
 188     *z2  = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(1, 1, 1, 1));
 189     *x3  = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(2, 2, 2, 2));
 190     *y3  = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(3, 3, 3, 3));
 191     *z3  = _mm_shuffle_ps(t3, t3, _MM_SHUFFLE(0, 0, 0, 0));
 192 }
 193
 194
 195 static gmx_inline void
 196 gmx_mm_load_shift_and_4rvec_broadcast_ps(const float * gmx_restrict xyz_shift,
 197                                          const float * gmx_restrict xyz,
 198                                          __m128 * gmx_restrict x1, __m128 * gmx_restrict y1, __m128 * gmx_restrict z1,
 199                                          __m128 * gmx_restrict x2, __m128 * gmx_restrict y2, __m128 * gmx_restrict z2,
 200                                          __m128 * gmx_restrict x3, __m128 * gmx_restrict y3, __m128 * gmx_restrict z3,
 201                                          __m128 * gmx_restrict x4, __m128 * gmx_restrict y4, __m128 * gmx_restrict z4)
 202 {
 203     __m128 tA, tB;
 204     __m128 t1, t2, t3, t4, t5, t6;
 205
 206     tA   = _mm_castpd_ps(_mm_load_sd((const double *)xyz_shift));
 207     tB   = _mm_load_ss(xyz_shift+2);
 208
 209     t1   = _mm_loadu_ps(xyz);
 210     t2   = _mm_loadu_ps(xyz+4);
 211     t3   = _mm_loadu_ps(xyz+8);
 212
 213     tA   = _mm_movelh_ps(tA, tB);
 214     t4   = _mm_shuffle_ps(tA, tA, _MM_SHUFFLE(0, 2, 1, 0));
 215     t5   = _mm_shuffle_ps(tA, tA, _MM_SHUFFLE(1, 0, 2, 1));
 216     t6   = _mm_shuffle_ps(tA, tA, _MM_SHUFFLE(2, 1, 0, 2));
 217
 218     t1   = _mm_add_ps(t1, t4);
 219     t2   = _mm_add_ps(t2, t5);
 220     t3   = _mm_add_ps(t3, t6);
 221
 222     *x1  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(0, 0, 0, 0));
 223     *y1  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(1, 1, 1, 1));
 224     *z1  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(2, 2, 2, 2));
 225     *x2  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(3, 3, 3, 3));
 226     *y2  = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(0, 0, 0, 0));
 227     *z2  = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(1, 1, 1, 1));
 228     *x3  = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(2, 2, 2, 2));
 229     *y3  = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(3, 3, 3, 3));
 230     *z3  = _mm_shuffle_ps(t3, t3, _MM_SHUFFLE(0, 0, 0, 0));
 231     *x4  = _mm_shuffle_ps(t3, t3, _MM_SHUFFLE(1, 1, 1, 1));
 232     *y4  = _mm_shuffle_ps(t3, t3, _MM_SHUFFLE(2, 2, 2, 2));
 233     *z4  = _mm_shuffle_ps(t3, t3, _MM_SHUFFLE(3, 3, 3, 3));
 234 }
 235
 236
 237 static void
 238 gmx_mm_load_1rvec_4ptr_swizzle_ps(const float * gmx_restrict ptrA,
 239                                   const float * gmx_restrict ptrB,
 240                                   const float * gmx_restrict ptrC,
 241                                   const float * gmx_restrict ptrD,
 242                                   __m128 *      gmx_restrict x1,
 243                                   __m128 *      gmx_restrict y1,
 244                                   __m128 *      gmx_restrict z1)
 245 {
 246     __m128 t1, t2, t3, t4, t5, t6, t7, t8;
 247     t1   = _mm_castpd_ps(_mm_load_sd((const double *)ptrA));
 248     t2   = _mm_castpd_ps(_mm_load_sd((const double *)ptrB));
 249     t3   = _mm_castpd_ps(_mm_load_sd((const double *)ptrC));
 250     t4   = _mm_castpd_ps(_mm_load_sd((const double *)ptrD));
 251     t5   = _mm_load_ss(ptrA+2);
 252     t6   = _mm_load_ss(ptrB+2);
 253     t7   = _mm_load_ss(ptrC+2);
 254     t8   = _mm_load_ss(ptrD+2);
 255     t1   = _mm_unpacklo_ps(t1, t2);
 256     t3   = _mm_unpacklo_ps(t3, t4);
 257     *x1  = _mm_movelh_ps(t1, t3);
 258     *y1  = _mm_movehl_ps(t3, t1);
 259     t5   = _mm_unpacklo_ps(t5, t6);
 260     t7   = _mm_unpacklo_ps(t7, t8);
 261     *z1  = _mm_movelh_ps(t5, t7);
 262 }
 263
 264
 265 static void
 266 gmx_mm_load_3rvec_4ptr_swizzle_ps(const float * gmx_restrict ptrA,
 267                                   const float * gmx_restrict ptrB,
 268                                   const float * gmx_restrict ptrC,
 269                                   const float * gmx_restrict ptrD,
 270                                   __m128 * gmx_restrict x1, __m128 * gmx_restrict y1, __m128 * gmx_restrict z1,
 271                                   __m128 * gmx_restrict x2, __m128 * gmx_restrict y2, __m128 * gmx_restrict z2,
 272                                   __m128 * gmx_restrict x3, __m128 * gmx_restrict y3, __m128 * gmx_restrict z3)
 273 {
 274     __m128 t1, t2, t3, t4;
 275     t1            = _mm_loadu_ps(ptrA);
 276     t2            = _mm_loadu_ps(ptrB);
 277     t3            = _mm_loadu_ps(ptrC);
 278     t4            = _mm_loadu_ps(ptrD);
 279     _MM_TRANSPOSE4_PS(t1, t2, t3, t4);
 280     *x1           = t1;
 281     *y1           = t2;
 282     *z1           = t3;
 283     *x2           = t4;
 284     t1            = _mm_loadu_ps(ptrA+4);
 285     t2            = _mm_loadu_ps(ptrB+4);
 286     t3            = _mm_loadu_ps(ptrC+4);
 287     t4            = _mm_loadu_ps(ptrD+4);
 288     _MM_TRANSPOSE4_PS(t1, t2, t3, t4);
 289     *y2           = t1;
 290     *z2           = t2;
 291     *x3           = t3;
 292     *y3           = t4;
 293     t1            = _mm_load_ss(ptrA+8);
 294     t2            = _mm_load_ss(ptrB+8);
 295     t3            = _mm_load_ss(ptrC+8);
 296     t4            = _mm_load_ss(ptrD+8);
 297     t1            = _mm_unpacklo_ps(t1, t3);
 298     t3            = _mm_unpacklo_ps(t2, t4);
 299     *z3           = _mm_unpacklo_ps(t1, t3);
 300 }
 301
 302
 303 static void
 304 gmx_mm_load_4rvec_4ptr_swizzle_ps(const float * gmx_restrict ptrA,
 305                                   const float * gmx_restrict ptrB,
 306                                   const float * gmx_restrict ptrC,
 307                                   const float * gmx_restrict ptrD,
 308                                   __m128 * gmx_restrict x1, __m128 * gmx_restrict y1, __m128 * gmx_restrict z1,
 309                                   __m128 * gmx_restrict x2, __m128 * gmx_restrict y2, __m128 * gmx_restrict z2,
 310                                   __m128 * gmx_restrict x3, __m128 * gmx_restrict y3, __m128 * gmx_restrict z3,
 311                                   __m128 * gmx_restrict x4, __m128 * gmx_restrict y4, __m128 * gmx_restrict z4)
 312 {
 313     __m128 t1, t2, t3, t4;
 314     t1            = _mm_loadu_ps(ptrA);
 315     t2            = _mm_loadu_ps(ptrB);
 316     t3            = _mm_loadu_ps(ptrC);
 317     t4            = _mm_loadu_ps(ptrD);
 318     _MM_TRANSPOSE4_PS(t1, t2, t3, t4);
 319     *x1           = t1;
 320     *y1           = t2;
 321     *z1           = t3;
 322     *x2           = t4;
 323     t1            = _mm_loadu_ps(ptrA+4);
 324     t2            = _mm_loadu_ps(ptrB+4);
 325     t3            = _mm_loadu_ps(ptrC+4);
 326     t4            = _mm_loadu_ps(ptrD+4);
 327     _MM_TRANSPOSE4_PS(t1, t2, t3, t4);
 328     *y2           = t1;
 329     *z2           = t2;
 330     *x3           = t3;
 331     *y3           = t4;
 332     t1            = _mm_loadu_ps(ptrA+8);
 333     t2            = _mm_loadu_ps(ptrB+8);
 334     t3            = _mm_loadu_ps(ptrC+8);
 335     t4            = _mm_loadu_ps(ptrD+8);
 336     _MM_TRANSPOSE4_PS(t1, t2, t3, t4);
 337     *z3           = t1;
 338     *x4           = t2;
 339     *y4           = t3;
 340     *z4           = t4;
 341 }
 342
 343
 344 static void
 345 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(float * gmx_restrict ptrA,
 346                                        float * gmx_restrict ptrB,
 347                                        float * gmx_restrict ptrC,
 348                                        float * gmx_restrict ptrD,
 349                                        __m128 x1, __m128 y1, __m128 z1)
 350 {
 351     __m128 t1, t2, t3, t4, t5, t6, t7, t8, t9, t10, t11, t12;
 352     t5          = _mm_unpacklo_ps(y1, z1);
 353     t6          = _mm_unpackhi_ps(y1, z1);
 354     t7          = _mm_shuffle_ps(x1, t5, _MM_SHUFFLE(1, 0, 0, 0));
 355     t8          = _mm_shuffle_ps(x1, t5, _MM_SHUFFLE(3, 2, 0, 1));
 356     t9          = _mm_shuffle_ps(x1, t6, _MM_SHUFFLE(1, 0, 0, 2));
 357     t10         = _mm_shuffle_ps(x1, t6, _MM_SHUFFLE(3, 2, 0, 3));
 358     t1          = _mm_load_ss(ptrA);
 359     t1          = _mm_loadh_pi(t1, (__m64 *)(ptrA+1));
 360     t1          = _mm_sub_ps(t1, t7);
 361     _mm_store_ss(ptrA, t1);
 362     _mm_storeh_pi((__m64 *)(ptrA+1), t1);
 363     t2          = _mm_load_ss(ptrB);
 364     t2          = _mm_loadh_pi(t2, (__m64 *)(ptrB+1));
 365     t2          = _mm_sub_ps(t2, t8);
 366     _mm_store_ss(ptrB, t2);
 367     _mm_storeh_pi((__m64 *)(ptrB+1), t2);
 368     t3          = _mm_load_ss(ptrC);
 369     t3          = _mm_loadh_pi(t3, (__m64 *)(ptrC+1));
 370     t3          = _mm_sub_ps(t3, t9);
 371     _mm_store_ss(ptrC, t3);
 372     _mm_storeh_pi((__m64 *)(ptrC+1), t3);
 373     t4          = _mm_load_ss(ptrD);
 374     t4          = _mm_loadh_pi(t4, (__m64 *)(ptrD+1));
 375     t4          = _mm_sub_ps(t4, t10);
 376     _mm_store_ss(ptrD, t4);
 377     _mm_storeh_pi((__m64 *)(ptrD+1), t4);
 378 }
 379
 380
 381
 382 #if defined (_MSC_VER) && defined(_M_IX86)
 383 /* Macro work-around since 32-bit MSVC cannot handle >3 xmm/ymm parameters */
 384 #define gmx_mm_decrement_3rvec_4ptr_swizzle_ps(ptrA, ptrB, ptrC, ptrD, \
 385                                                _x1, _y1, _z1, _x2, _y2, _z2, _x3, _y3, _z3) \
 386     { \
 387         __m128 _t1, _t2, _t3, _t4, _t5, _t6, _t7, _t8, _t9, _t10; \
 388         __m128 _t11, _t12, _t13, _t14, _t15, _t16, _t17, _t18, _t19; \
 389         __m128 _t20, _t21, _t22, _t23, _t24, _t25; \
 390         _t13         = _mm_unpackhi_ps(_x1, _y1); \
 391         _x1          = _mm_unpacklo_ps(_x1, _y1); \
 392         _t14         = _mm_unpackhi_ps(_z1, _x2); \
 393         _z1          = _mm_unpacklo_ps(_z1, _x2); \
 394         _t15         = _mm_unpackhi_ps(_y2, _z2); \
 395         _y2          = _mm_unpacklo_ps(_y2, _z2); \
 396         _t16         = _mm_unpackhi_ps(_x3, _y3); \
 397         _x3          = _mm_unpacklo_ps(_x3, _y3); \
 398         _t17         = _mm_shuffle_ps(_z3, _z3, _MM_SHUFFLE(0, 0, 0, 1)); \
 399         _t18         = _mm_movehl_ps(_z3, _z3); \
 400         _t19         = _mm_shuffle_ps(_t18, _t18, _MM_SHUFFLE(0, 0, 0, 1)); \
 401         _t20         = _mm_movelh_ps(_x1, _z1); \
 402         _t21         = _mm_movehl_ps(_z1, _x1); \
 403         _t22         = _mm_movelh_ps(_t13, _t14); \
 404         _t14         = _mm_movehl_ps(_t14, _t13); \
 405         _t23         = _mm_movelh_ps(_y2, _x3); \
 406         _t24         = _mm_movehl_ps(_x3, _y2); \
 407         _t25         = _mm_movelh_ps(_t15, _t16); \
 408         _t16         = _mm_movehl_ps(_t16, _t15); \
 409         _t1          = _mm_loadu_ps(ptrA); \
 410         _t2          = _mm_loadu_ps(ptrA+4); \
 411         _t3          = _mm_load_ss(ptrA+8); \
 412         _t1          = _mm_sub_ps(_t1, _t20); \
 413         _t2          = _mm_sub_ps(_t2, _t23); \
 414         _t3          = _mm_sub_ss(_t3, _z3); \
 415         _mm_storeu_ps(ptrA, _t1); \
 416         _mm_storeu_ps(ptrA+4, _t2); \
 417         _mm_store_ss(ptrA+8, _t3); \
 418         _t4          = _mm_loadu_ps(ptrB); \
 419         _t5          = _mm_loadu_ps(ptrB+4); \
 420         _t6          = _mm_load_ss(ptrB+8); \
 421         _t4          = _mm_sub_ps(_t4, _t21); \
 422         _t5          = _mm_sub_ps(_t5, _t24); \
 423         _t6          = _mm_sub_ss(_t6, _t17); \
 424         _mm_storeu_ps(ptrB, _t4); \
 425         _mm_storeu_ps(ptrB+4, _t5); \
 426         _mm_store_ss(ptrB+8, _t6); \
 427         _t7          = _mm_loadu_ps(ptrC); \
 428         _t8          = _mm_loadu_ps(ptrC+4); \
 429         _t9          = _mm_load_ss(ptrC+8); \
 430         _t7          = _mm_sub_ps(_t7, _t22); \
 431         _t8          = _mm_sub_ps(_t8, _t25); \
 432         _t9          = _mm_sub_ss(_t9, _t18); \
 433         _mm_storeu_ps(ptrC, _t7); \
 434         _mm_storeu_ps(ptrC+4, _t8); \
 435         _mm_store_ss(ptrC+8, _t9); \
 436         _t10         = _mm_loadu_ps(ptrD); \
 437         _t11         = _mm_loadu_ps(ptrD+4); \
 438         _t12         = _mm_load_ss(ptrD+8); \
 439         _t10         = _mm_sub_ps(_t10, _t14); \
 440         _t11         = _mm_sub_ps(_t11, _t16); \
 441         _t12         = _mm_sub_ss(_t12, _t19); \
 442         _mm_storeu_ps(ptrD, _t10); \
 443         _mm_storeu_ps(ptrD+4, _t11); \
 444         _mm_store_ss(ptrD+8, _t12); \
 445     }
 446 #else
 447 /* Real function for sane compilers */
 448 static void
 449 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(float * gmx_restrict ptrA, float * gmx_restrict ptrB,
 450                                        float * gmx_restrict ptrC, float * gmx_restrict ptrD,
 451                                        __m128 x1, __m128 y1, __m128 z1,
 452                                        __m128 x2, __m128 y2, __m128 z2,
 453                                        __m128 x3, __m128 y3, __m128 z3)
 454 {
 455     __m128 t1, t2, t3, t4, t5, t6, t7, t8, t9, t10;
 456     __m128 t11, t12, t13, t14, t15, t16, t17, t18, t19;
 457     __m128 t20, t21, t22, t23, t24, t25;
 458
 459     t13         = _mm_unpackhi_ps(x1, y1);
 460     x1          = _mm_unpacklo_ps(x1, y1);
 461     t14         = _mm_unpackhi_ps(z1, x2);
 462     z1          = _mm_unpacklo_ps(z1, x2);
 463     t15         = _mm_unpackhi_ps(y2, z2);
 464     y2          = _mm_unpacklo_ps(y2, z2);
 465     t16         = _mm_unpackhi_ps(x3, y3);
 466     x3          = _mm_unpacklo_ps(x3, y3);
 467     t17         = _mm_shuffle_ps(z3, z3, _MM_SHUFFLE(0, 0, 0, 1));
 468     t18         = _mm_movehl_ps(z3, z3);
 469     t19         = _mm_shuffle_ps(t18, t18, _MM_SHUFFLE(0, 0, 0, 1));
 470     t20         = _mm_movelh_ps(x1, z1);
 471     t21         = _mm_movehl_ps(z1, x1);
 472     t22         = _mm_movelh_ps(t13, t14);
 473     t14         = _mm_movehl_ps(t14, t13);
 474     t23         = _mm_movelh_ps(y2, x3);
 475     t24         = _mm_movehl_ps(x3, y2);
 476     t25         = _mm_movelh_ps(t15, t16);
 477     t16         = _mm_movehl_ps(t16, t15);
 478     t1          = _mm_loadu_ps(ptrA);
 479     t2          = _mm_loadu_ps(ptrA+4);
 480     t3          = _mm_load_ss(ptrA+8);
 481     t1          = _mm_sub_ps(t1, t20);
 482     t2          = _mm_sub_ps(t2, t23);
 483     t3          = _mm_sub_ss(t3, z3);
 484     _mm_storeu_ps(ptrA, t1);
 485     _mm_storeu_ps(ptrA+4, t2);
 486     _mm_store_ss(ptrA+8, t3);
 487     t4          = _mm_loadu_ps(ptrB);
 488     t5          = _mm_loadu_ps(ptrB+4);
 489     t6          = _mm_load_ss(ptrB+8);
 490     t4          = _mm_sub_ps(t4, t21);
 491     t5          = _mm_sub_ps(t5, t24);
 492     t6          = _mm_sub_ss(t6, t17);
 493     _mm_storeu_ps(ptrB, t4);
 494     _mm_storeu_ps(ptrB+4, t5);
 495     _mm_store_ss(ptrB+8, t6);
 496     t7          = _mm_loadu_ps(ptrC);
 497     t8          = _mm_loadu_ps(ptrC+4);
 498     t9          = _mm_load_ss(ptrC+8);
 499     t7          = _mm_sub_ps(t7, t22);
 500     t8          = _mm_sub_ps(t8, t25);
 501     t9          = _mm_sub_ss(t9, t18);
 502     _mm_storeu_ps(ptrC, t7);
 503     _mm_storeu_ps(ptrC+4, t8);
 504     _mm_store_ss(ptrC+8, t9);
 505     t10         = _mm_loadu_ps(ptrD);
 506     t11         = _mm_loadu_ps(ptrD+4);
 507     t12         = _mm_load_ss(ptrD+8);
 508     t10         = _mm_sub_ps(t10, t14);
 509     t11         = _mm_sub_ps(t11, t16);
 510     t12         = _mm_sub_ss(t12, t19);
 511     _mm_storeu_ps(ptrD, t10);
 512     _mm_storeu_ps(ptrD+4, t11);
 513     _mm_store_ss(ptrD+8, t12);
 514 }
 515 #endif
 516
 517
 518 #if defined (_MSC_VER) && defined(_M_IX86)
 519 /* Macro work-around since 32-bit MSVC cannot handle >3 xmm/ymm parameters */
 520 #define gmx_mm_decrement_4rvec_4ptr_swizzle_ps(ptrA, ptrB, ptrC, ptrD, \
 521                                                _x1, _y1, _z1, _x2, _y2, _z2, _x3, _y3, _z3, _x4, _y4, _z4) \
 522     { \
 523         __m128 _t1, _t2, _t3, _t4, _t5, _t6, _t7, _t8, _t9, _t10, _t11; \
 524         __m128 _t12, _t13, _t14, _t15, _t16, _t17, _t18, _t19, _t20, _t21, _t22; \
 525         __m128 _t23, _t24; \
 526         _t13         = _mm_unpackhi_ps(_x1, _y1); \
 527         _x1          = _mm_unpacklo_ps(_x1, _y1); \
 528         _t14         = _mm_unpackhi_ps(_z1, _x2); \
 529         _z1          = _mm_unpacklo_ps(_z1, _x2); \
 530         _t15         = _mm_unpackhi_ps(_y2, _z2); \
 531         _y2          = _mm_unpacklo_ps(_y2, _z2); \
 532         _t16         = _mm_unpackhi_ps(_x3, _y3); \
 533         _x3          = _mm_unpacklo_ps(_x3, _y3); \
 534         _t17         = _mm_unpackhi_ps(_z3, _x4); \
 535         _z3          = _mm_unpacklo_ps(_z3, _x4); \
 536         _t18         = _mm_unpackhi_ps(_y4, _z4); \
 537         _y4          = _mm_unpacklo_ps(_y4, _z4); \
 538         _t19         = _mm_movelh_ps(_x1, _z1); \
 539         _z1          = _mm_movehl_ps(_z1, _x1); \
 540         _t20         = _mm_movelh_ps(_t13, _t14); \
 541         _t14         = _mm_movehl_ps(_t14, _t13); \
 542         _t21         = _mm_movelh_ps(_y2, _x3); \
 543         _x3          = _mm_movehl_ps(_x3, _y2); \
 544         _t22         = _mm_movelh_ps(_t15, _t16); \
 545         _t16         = _mm_movehl_ps(_t16, _t15); \
 546         _t23         = _mm_movelh_ps(_z3, _y4); \
 547         _y4          = _mm_movehl_ps(_y4, _z3); \
 548         _t24         = _mm_movelh_ps(_t17, _t18); \
 549         _t18         = _mm_movehl_ps(_t18, _t17); \
 550         _t1          = _mm_loadu_ps(ptrA); \
 551         _t2          = _mm_loadu_ps(ptrA+4); \
 552         _t3          = _mm_loadu_ps(ptrA+8); \
 553         _t1          = _mm_sub_ps(_t1, _t19); \
 554         _t2          = _mm_sub_ps(_t2, _t21); \
 555         _t3          = _mm_sub_ps(_t3, _t23); \
 556         _mm_storeu_ps(ptrA, _t1); \
 557         _mm_storeu_ps(ptrA+4, _t2); \
 558         _mm_storeu_ps(ptrA+8, _t3); \
 559         _t4          = _mm_loadu_ps(ptrB); \
 560         _t5          = _mm_loadu_ps(ptrB+4); \
 561         _t6          = _mm_loadu_ps(ptrB+8); \
 562         _t4          = _mm_sub_ps(_t4, _z1); \
 563         _t5          = _mm_sub_ps(_t5, _x3); \
 564         _t6          = _mm_sub_ps(_t6, _y4); \
 565         _mm_storeu_ps(ptrB, _t4); \
 566         _mm_storeu_ps(ptrB+4, _t5); \
 567         _mm_storeu_ps(ptrB+8, _t6); \
 568         _t7          = _mm_loadu_ps(ptrC); \
 569         _t8          = _mm_loadu_ps(ptrC+4); \
 570         _t9          = _mm_loadu_ps(ptrC+8); \
 571         _t7          = _mm_sub_ps(_t7, _t20); \
 572         _t8          = _mm_sub_ps(_t8, _t22); \
 573         _t9          = _mm_sub_ps(_t9, _t24); \
 574         _mm_storeu_ps(ptrC, _t7); \
 575         _mm_storeu_ps(ptrC+4, _t8); \
 576         _mm_storeu_ps(ptrC+8, _t9); \
 577         _t10         = _mm_loadu_ps(ptrD); \
 578         _t11         = _mm_loadu_ps(ptrD+4); \
 579         _t12         = _mm_loadu_ps(ptrD+8); \
 580         _t10         = _mm_sub_ps(_t10, _t14); \
 581         _t11         = _mm_sub_ps(_t11, _t16); \
 582         _t12         = _mm_sub_ps(_t12, _t18); \
 583         _mm_storeu_ps(ptrD, _t10); \
 584         _mm_storeu_ps(ptrD+4, _t11); \
 585         _mm_storeu_ps(ptrD+8, _t12); \
 586     }
 587 #else
 588 /* Real function for sane compilers */
 589 static void
 590 gmx_mm_decrement_4rvec_4ptr_swizzle_ps(float * gmx_restrict ptrA, float * gmx_restrict ptrB,
 591                                        float * gmx_restrict ptrC, float * gmx_restrict ptrD,
 592                                        __m128 x1, __m128 y1, __m128 z1,
 593                                        __m128 x2, __m128 y2, __m128 z2,
 594                                        __m128 x3, __m128 y3, __m128 z3,
 595                                        __m128 x4, __m128 y4, __m128 z4)
 596 {
 597     __m128 t1, t2, t3, t4, t5, t6, t7, t8, t9, t10, t11;
 598     __m128 t12, t13, t14, t15, t16, t17, t18, t19, t20, t21, t22;
 599     __m128 t23, t24;
 600     t13         = _mm_unpackhi_ps(x1, y1);
 601     x1          = _mm_unpacklo_ps(x1, y1);
 602     t14         = _mm_unpackhi_ps(z1, x2);
 603     z1          = _mm_unpacklo_ps(z1, x2);
 604     t15         = _mm_unpackhi_ps(y2, z2);
 605     y2          = _mm_unpacklo_ps(y2, z2);
 606     t16         = _mm_unpackhi_ps(x3, y3);
 607     x3          = _mm_unpacklo_ps(x3, y3);
 608     t17         = _mm_unpackhi_ps(z3, x4);
 609     z3          = _mm_unpacklo_ps(z3, x4);
 610     t18         = _mm_unpackhi_ps(y4, z4);
 611     y4          = _mm_unpacklo_ps(y4, z4);
 612     t19         = _mm_movelh_ps(x1, z1);
 613     z1          = _mm_movehl_ps(z1, x1);
 614     t20         = _mm_movelh_ps(t13, t14);
 615     t14         = _mm_movehl_ps(t14, t13);
 616     t21         = _mm_movelh_ps(y2, x3);
 617     x3          = _mm_movehl_ps(x3, y2);
 618     t22         = _mm_movelh_ps(t15, t16);
 619     t16         = _mm_movehl_ps(t16, t15);
 620     t23         = _mm_movelh_ps(z3, y4);
 621     y4          = _mm_movehl_ps(y4, z3);
 622     t24         = _mm_movelh_ps(t17, t18);
 623     t18         = _mm_movehl_ps(t18, t17);
 624     t1          = _mm_loadu_ps(ptrA);
 625     t2          = _mm_loadu_ps(ptrA+4);
 626     t3          = _mm_loadu_ps(ptrA+8);
 627     t1          = _mm_sub_ps(t1, t19);
 628     t2          = _mm_sub_ps(t2, t21);
 629     t3          = _mm_sub_ps(t3, t23);
 630     _mm_storeu_ps(ptrA, t1);
 631     _mm_storeu_ps(ptrA+4, t2);
 632     _mm_storeu_ps(ptrA+8, t3);
 633     t4          = _mm_loadu_ps(ptrB);
 634     t5          = _mm_loadu_ps(ptrB+4);
 635     t6          = _mm_loadu_ps(ptrB+8);
 636     t4          = _mm_sub_ps(t4, z1);
 637     t5          = _mm_sub_ps(t5, x3);
 638     t6          = _mm_sub_ps(t6, y4);
 639     _mm_storeu_ps(ptrB, t4);
 640     _mm_storeu_ps(ptrB+4, t5);
 641     _mm_storeu_ps(ptrB+8, t6);
 642     t7          = _mm_loadu_ps(ptrC);
 643     t8          = _mm_loadu_ps(ptrC+4);
 644     t9          = _mm_loadu_ps(ptrC+8);
 645     t7          = _mm_sub_ps(t7, t20);
 646     t8          = _mm_sub_ps(t8, t22);
 647     t9          = _mm_sub_ps(t9, t24);
 648     _mm_storeu_ps(ptrC, t7);
 649     _mm_storeu_ps(ptrC+4, t8);
 650     _mm_storeu_ps(ptrC+8, t9);
 651     t10         = _mm_loadu_ps(ptrD);
 652     t11         = _mm_loadu_ps(ptrD+4);
 653     t12         = _mm_loadu_ps(ptrD+8);
 654     t10         = _mm_sub_ps(t10, t14);
 655     t11         = _mm_sub_ps(t11, t16);
 656     t12         = _mm_sub_ps(t12, t18);
 657     _mm_storeu_ps(ptrD, t10);
 658     _mm_storeu_ps(ptrD+4, t11);
 659     _mm_storeu_ps(ptrD+8, t12);
 660 }
 661 #endif
 662
 663
 664 static gmx_inline void
 665 gmx_mm_update_iforce_1atom_swizzle_ps(__m128 fix1, __m128 fiy1, __m128 fiz1,
 666                                       float * gmx_restrict fptr,
 667                                       float * gmx_restrict fshiftptr)
 668 {
 669     __m128 t1, t2, t3;
 670
 671     /* transpose data */
 672     t1 = fix1;
 673     _MM_TRANSPOSE4_PS(fix1, t1, fiy1, fiz1);
 674     fix1 = _mm_add_ps(_mm_add_ps(fix1, t1), _mm_add_ps(fiy1, fiz1));
 675
 676     t2 = _mm_load_ss(fptr);
 677     t2 = _mm_loadh_pi(t2, (__m64 *)(fptr+1));
 678     t3 = _mm_load_ss(fshiftptr);
 679     t3 = _mm_loadh_pi(t3, (__m64 *)(fshiftptr+1));
 680
 681     t2 = _mm_add_ps(t2, fix1);
 682     t3 = _mm_add_ps(t3, fix1);
 683
 684     _mm_store_ss(fptr, t2);
 685     _mm_storeh_pi((__m64 *)(fptr+1), t2);
 686     _mm_store_ss(fshiftptr, t3);
 687     _mm_storeh_pi((__m64 *)(fshiftptr+1), t3);
 688 }
 689
 690 #if defined (_MSC_VER) && defined(_M_IX86)
 691 /* Macro work-around since 32-bit MSVC cannot handle >3 xmm/ymm parameters */
 692 #define gmx_mm_update_iforce_3atom_swizzle_ps(fix1, fiy1, fiz1, fix2, fiy2, fiz2, fix3, fiy3, fiz3, \
 693                                               fptr, fshiftptr) \
 694     { \
 695         __m128 _t1, _t2, _t3, _t4; \
 696 \
 697         _MM_TRANSPOSE4_PS(fix1, fiy1, fiz1, fix2); \
 698         _MM_TRANSPOSE4_PS(fiy2, fiz2, fix3, fiy3); \
 699         _t2   = _mm_movehl_ps(_mm_setzero_ps(), fiz3); \
 700         _t1   = _mm_shuffle_ps(fiz3, fiz3, _MM_SHUFFLE(0, 0, 0, 1)); \
 701         _t3   = _mm_shuffle_ps(_t2, _t2, _MM_SHUFFLE(0, 0, 0, 1)); \
 702         fix1  = _mm_add_ps(_mm_add_ps(fix1, fiy1), _mm_add_ps(fiz1, fix2)); \
 703         fiy2  = _mm_add_ps(_mm_add_ps(fiy2, fiz2), _mm_add_ps(fix3, fiy3)); \
 704         fiz3  = _mm_add_ss(_mm_add_ps(fiz3, _t1), _mm_add_ps(_t2, _t3)); \
 705         _mm_storeu_ps(fptr,  _mm_add_ps(fix1, _mm_loadu_ps(fptr)  )); \
 706         _mm_storeu_ps(fptr+4, _mm_add_ps(fiy2, _mm_loadu_ps(fptr+4))); \
 707         _mm_store_ss (fptr+8, _mm_add_ss(fiz3, _mm_load_ss(fptr+8) )); \
 708         _t4 = _mm_load_ss(fshiftptr+2); \
 709         _t4 = _mm_loadh_pi(_t4, (__m64 *)(fshiftptr)); \
 710         _t1 = _mm_shuffle_ps(fiz3, fix1, _MM_SHUFFLE(1, 0, 0, 0)); \
 711         _t2 = _mm_shuffle_ps(fix1, fiy2, _MM_SHUFFLE(3, 2, 2, 2)); \
 712         _t3 = _mm_shuffle_ps(fiy2, fix1, _MM_SHUFFLE(3, 3, 0, 1)); \
 713         _t3 = _mm_shuffle_ps(_t3, _t3, _MM_SHUFFLE(1, 2, 0, 0)); \
 714         _t1 = _mm_add_ps(_t1, _t2); \
 715         _t3 = _mm_add_ps(_t3, _t4); \
 716         _t1 = _mm_add_ps(_t1, _t3); \
 717         _mm_store_ss(fshiftptr+2, _t1); \
 718         _mm_storeh_pi((__m64 *)(fshiftptr), _t1); \
 719     }
 720 #else
 721 /* Real function for sane compilers */
 722 static gmx_inline void
 723 gmx_mm_update_iforce_3atom_swizzle_ps(__m128 fix1, __m128 fiy1, __m128 fiz1,
 724                                       __m128 fix2, __m128 fiy2, __m128 fiz2,
 725                                       __m128 fix3, __m128 fiy3, __m128 fiz3,
 726                                       float * gmx_restrict fptr,
 727                                       float * gmx_restrict fshiftptr)
 728 {
 729     __m128 t1, t2, t3, t4;
 730
 731     /* transpose data */
 732     _MM_TRANSPOSE4_PS(fix1, fiy1, fiz1, fix2);
 733     _MM_TRANSPOSE4_PS(fiy2, fiz2, fix3, fiy3);
 734     t2   = _mm_movehl_ps(_mm_setzero_ps(), fiz3);
 735     t1   = _mm_shuffle_ps(fiz3, fiz3, _MM_SHUFFLE(0, 0, 0, 1));
 736     t3   = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(0, 0, 0, 1));
 737
 738     fix1 = _mm_add_ps(_mm_add_ps(fix1, fiy1), _mm_add_ps(fiz1, fix2));
 739     fiy2 = _mm_add_ps(_mm_add_ps(fiy2, fiz2), _mm_add_ps(fix3, fiy3));
 740     fiz3 = _mm_add_ss(_mm_add_ps(fiz3, t1), _mm_add_ps(t2, t3));
 741
 742     _mm_storeu_ps(fptr,  _mm_add_ps(fix1, _mm_loadu_ps(fptr)  ));
 743     _mm_storeu_ps(fptr+4, _mm_add_ps(fiy2, _mm_loadu_ps(fptr+4)));
 744     _mm_store_ss (fptr+8, _mm_add_ss(fiz3, _mm_load_ss(fptr+8) ));
 745
 746     t4 = _mm_load_ss(fshiftptr+2);
 747     t4 = _mm_loadh_pi(t4, (__m64 *)(fshiftptr));
 748
 749     t1 = _mm_shuffle_ps(fiz3, fix1, _MM_SHUFFLE(1, 0, 0, 0)); /* fiy1 fix1  -   fiz3 */
 750     t2 = _mm_shuffle_ps(fix1, fiy2, _MM_SHUFFLE(3, 2, 2, 2)); /* fiy3 fix3  -   fiz1 */
 751     t3 = _mm_shuffle_ps(fiy2, fix1, _MM_SHUFFLE(3, 3, 0, 1)); /* fix2 fix2 fiy2 fiz2 */
 752     t3 = _mm_shuffle_ps(t3, t3, _MM_SHUFFLE(1, 2, 0, 0));     /* fiy2 fix2  -   fiz2 */
 753
 754     t1 = _mm_add_ps(t1, t2);
 755     t3 = _mm_add_ps(t3, t4);
 756     t1 = _mm_add_ps(t1, t3); /* y x - z */
 757
 758     _mm_store_ss(fshiftptr+2, t1);
 759     _mm_storeh_pi((__m64 *)(fshiftptr), t1);
 760 }
 761 #endif
 762
 763 #if defined (_MSC_VER) && defined(_M_IX86)
 764 /* Macro work-around since 32-bit MSVC cannot handle >3 xmm/ymm parameters */
 765 #define gmx_mm_update_iforce_4atom_swizzle_ps(fix1, fiy1, fiz1, fix2, fiy2, fiz2, fix3, fiy3, fiz3, fix4, fiy4, fiz4, \
 766                                               fptr, fshiftptr) \
 767     { \
 768         __m128 _t1, _t2, _t3, _t4, _t5; \
 769         _MM_TRANSPOSE4_PS(fix1, fiy1, fiz1, fix2); \
 770         _MM_TRANSPOSE4_PS(fiy2, fiz2, fix3, fiy3); \
 771         _MM_TRANSPOSE4_PS(fiz3, fix4, fiy4, fiz4); \
 772         fix1 = _mm_add_ps(_mm_add_ps(fix1, fiy1), _mm_add_ps(fiz1, fix2)); \
 773         fiy2 = _mm_add_ps(_mm_add_ps(fiy2, fiz2), _mm_add_ps(fix3, fiy3)); \
 774         fiz3 = _mm_add_ps(_mm_add_ps(fiz3, fix4), _mm_add_ps(fiy4, fiz4)); \
 775         _mm_storeu_ps(fptr,  _mm_add_ps(fix1, _mm_loadu_ps(fptr)  )); \
 776         _mm_storeu_ps(fptr+4, _mm_add_ps(fiy2, _mm_loadu_ps(fptr+4))); \
 777         _mm_storeu_ps(fptr+8, _mm_add_ps(fiz3, _mm_loadu_ps(fptr+8))); \
 778         _t5 = _mm_load_ss(fshiftptr+2); \
 779         _t5 = _mm_loadh_pi(_t5, (__m64 *)(fshiftptr)); \
 780         _t1 = _mm_shuffle_ps(fix1, fix1, _MM_SHUFFLE(1, 0, 2, 2)); \
 781         _t2 = _mm_shuffle_ps(fiy2, fiy2, _MM_SHUFFLE(3, 2, 1, 1)); \
 782         _t3 = _mm_shuffle_ps(fiz3, fiz3, _MM_SHUFFLE(2, 1, 0, 0)); \
 783         _t4 = _mm_shuffle_ps(fix1, fiy2, _MM_SHUFFLE(0, 0, 3, 3)); \
 784         _t4 = _mm_shuffle_ps(fiz3, _t4, _MM_SHUFFLE(2, 0, 3, 3)); \
 785         _t1 = _mm_add_ps(_t1, _t2); \
 786         _t3 = _mm_add_ps(_t3, _t4); \
 787         _t1 = _mm_add_ps(_t1, _t3); \
 788         _t5 = _mm_add_ps(_t5, _t1); \
 789         _mm_store_ss(fshiftptr+2, _t5); \
 790         _mm_storeh_pi((__m64 *)(fshiftptr), _t5); \
 791     }
 792 #else
 793 /* Real function for sane compilers */
 794 static gmx_inline void
 795 gmx_mm_update_iforce_4atom_swizzle_ps(__m128 fix1, __m128 fiy1, __m128 fiz1,
 796                                       __m128 fix2, __m128 fiy2, __m128 fiz2,
 797                                       __m128 fix3, __m128 fiy3, __m128 fiz3,
 798                                       __m128 fix4, __m128 fiy4, __m128 fiz4,
 799                                       float * gmx_restrict fptr,
 800                                       float * gmx_restrict fshiftptr)
 801 {
 802     __m128 t1, t2, t3, t4, t5;
 803
 804     /* transpose data */
 805     _MM_TRANSPOSE4_PS(fix1, fiy1, fiz1, fix2);
 806     _MM_TRANSPOSE4_PS(fiy2, fiz2, fix3, fiy3);
 807     _MM_TRANSPOSE4_PS(fiz3, fix4, fiy4, fiz4);
 808
 809     fix1 = _mm_add_ps(_mm_add_ps(fix1, fiy1), _mm_add_ps(fiz1, fix2));
 810     fiy2 = _mm_add_ps(_mm_add_ps(fiy2, fiz2), _mm_add_ps(fix3, fiy3));
 811     fiz3 = _mm_add_ps(_mm_add_ps(fiz3, fix4), _mm_add_ps(fiy4, fiz4));
 812
 813     _mm_storeu_ps(fptr,  _mm_add_ps(fix1, _mm_loadu_ps(fptr)  ));
 814     _mm_storeu_ps(fptr+4, _mm_add_ps(fiy2, _mm_loadu_ps(fptr+4)));
 815     _mm_storeu_ps(fptr+8, _mm_add_ps(fiz3, _mm_loadu_ps(fptr+8)));
 816
 817     t5 = _mm_load_ss(fshiftptr+2);
 818     t5 = _mm_loadh_pi(t5, (__m64 *)(fshiftptr));
 819
 820     t1 = _mm_shuffle_ps(fix1, fix1, _MM_SHUFFLE(1, 0, 2, 2));
 821     t2 = _mm_shuffle_ps(fiy2, fiy2, _MM_SHUFFLE(3, 2, 1, 1));
 822     t3 = _mm_shuffle_ps(fiz3, fiz3, _MM_SHUFFLE(2, 1, 0, 0));
 823     t4 = _mm_shuffle_ps(fix1, fiy2, _MM_SHUFFLE(0, 0, 3, 3));
 824     t4 = _mm_shuffle_ps(fiz3, t4, _MM_SHUFFLE(2, 0, 3, 3));
 825
 826     t1 = _mm_add_ps(t1, t2);
 827     t3 = _mm_add_ps(t3, t4);
 828     t1 = _mm_add_ps(t1, t3);
 829     t5 = _mm_add_ps(t5, t1);
 830
 831     _mm_store_ss(fshiftptr+2, t5);
 832     _mm_storeh_pi((__m64 *)(fshiftptr), t5);
 833 }
 834 #endif
 835
 836
 837 static void
 838 gmx_mm_update_1pot_ps(__m128 pot1, float * gmx_restrict ptrA)
 839 {
 840     pot1 = _mm_add_ps(pot1, _mm_movehl_ps(_mm_setzero_ps(), pot1));
 841     pot1 = _mm_add_ps(pot1, _mm_shuffle_ps(pot1, pot1, _MM_SHUFFLE(0, 0, 0, 1)));
 842     _mm_store_ss(ptrA, _mm_add_ss(pot1, _mm_load_ss(ptrA)));
 843 }
 844
 845 static void
 846 gmx_mm_update_2pot_ps(__m128 pot1, float * gmx_restrict ptrA,
 847                       __m128 pot2, float * gmx_restrict ptrB)
 848 {
 849     __m128 t1, t2;
 850     t1   = _mm_movehl_ps(pot2, pot1);
 851     t2   = _mm_movelh_ps(pot1, pot2);
 852     t1   = _mm_add_ps(t1, t2);
 853     t2   = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(3, 3, 1, 1));
 854     pot1 = _mm_add_ps(t1, t2);
 855     pot2 = _mm_movehl_ps(t2, pot1);
 856     _mm_store_ss(ptrA, _mm_add_ss(pot1, _mm_load_ss(ptrA)));
 857     _mm_store_ss(ptrB, _mm_add_ss(pot2, _mm_load_ss(ptrB)));
 858 }
 859
 860
 861 #endif /* _kernelutil_x86_sse2_single_h_ */