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) 2012,2013,2014, 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 #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 #define gmx_mm_castsi128_ps   _mm_castsi128_ps
  43
  44 #define gmx_mm_extract_epi32(x, imm) _mm_cvtsi128_si32(_mm_srli_si128((x), 4 * (imm)))
  45
  46
  47 /* Normal sum of four xmm registers */
  48 #define gmx_mm_sum4_ps(t0, t1, t2, t3)  _mm_add_ps(_mm_add_ps(t0, t1), _mm_add_ps(t2, t3))
  49
  50 static gmx_inline __m128
  51 gmx_mm_calc_rsq_ps(__m128 dx, __m128 dy, __m128 dz)
  52 {
  53     return _mm_add_ps( _mm_add_ps( _mm_mul_ps(dx, dx), _mm_mul_ps(dy, dy) ), _mm_mul_ps(dz, dz) );
  54 }
  55
  56 static int
  57 gmx_mm_any_lt(__m128 a, __m128 b)
  58 {
  59     return _mm_movemask_ps(_mm_cmplt_ps(a, b));
  60 }
  61
  62 /* Load a single value from 1-4 places, merge into xmm register */
  63
  64 static __m128
  65 gmx_mm_load_4real_swizzle_ps(const float * gmx_restrict ptrA,
  66                              const float * gmx_restrict ptrB,
  67                              const float * gmx_restrict ptrC,
  68                              const float * gmx_restrict ptrD)
  69 {
  70     __m128 t1, t2;
  71
  72     t1 = _mm_unpacklo_ps(_mm_load_ss(ptrA), _mm_load_ss(ptrC));
  73     t2 = _mm_unpacklo_ps(_mm_load_ss(ptrB), _mm_load_ss(ptrD));
  74     return _mm_unpacklo_ps(t1, t2);
  75 }
  76
  77 static void
  78 gmx_mm_store_4real_swizzle_ps(float * gmx_restrict ptrA,
  79                               float * gmx_restrict ptrB,
  80                               float * gmx_restrict ptrC,
  81                               float * gmx_restrict ptrD,
  82                               __m128               xmm1)
  83 {
  84     __m128 t2, t3, t4;
  85
  86     t3       = _mm_movehl_ps(_mm_setzero_ps(), xmm1);
  87     t2       = _mm_shuffle_ps(xmm1, xmm1, _MM_SHUFFLE(1, 1, 1, 1));
  88     t4       = _mm_shuffle_ps(t3, t3, _MM_SHUFFLE(1, 1, 1, 1));
  89     _mm_store_ss(ptrA, xmm1);
  90     _mm_store_ss(ptrB, t2);
  91     _mm_store_ss(ptrC, t3);
  92     _mm_store_ss(ptrD, t4);
  93 }
  94
  95 /* Similar to store, but increments value in memory */
  96 static void
  97 gmx_mm_increment_4real_swizzle_ps(float * gmx_restrict ptrA,
  98                                   float * gmx_restrict ptrB,
  99                                   float * gmx_restrict ptrC,
 100                                   float * gmx_restrict ptrD, __m128 xmm1)
 101 {
 102     __m128 tmp;
 103
 104     tmp = gmx_mm_load_4real_swizzle_ps(ptrA, ptrB, ptrC, ptrD);
 105     tmp = _mm_add_ps(tmp, xmm1);
 106     gmx_mm_store_4real_swizzle_ps(ptrA, ptrB, ptrC, ptrD, tmp);
 107 }
 108
 109
 110 static void
 111 gmx_mm_load_4pair_swizzle_ps(const float * gmx_restrict p1,
 112                              const float * gmx_restrict p2,
 113                              const float * gmx_restrict p3,
 114                              const float * gmx_restrict p4,
 115                              __m128 * gmx_restrict      c6,
 116                              __m128 * gmx_restrict      c12)
 117 {
 118     __m128 t1, t2, t3, t4;
 119
 120     t1   = _mm_loadl_pi(_mm_setzero_ps(), (__m64 *)p1);   /* - - c12a  c6a */
 121     t2   = _mm_loadl_pi(_mm_setzero_ps(), (__m64 *)p2);   /* - - c12b  c6b */
 122     t3   = _mm_loadl_pi(_mm_setzero_ps(), (__m64 *)p3);   /* - - c12c  c6c */
 123     t4   = _mm_loadl_pi(_mm_setzero_ps(), (__m64 *)p4);   /* - - c12d  c6d */
 124     t1   = _mm_unpacklo_ps(t1, t2);
 125     t2   = _mm_unpacklo_ps(t3, t4);
 126     *c6  = _mm_movelh_ps(t1, t2);
 127     *c12 = _mm_movehl_ps(t2, t1);
 128 }
 129
 130 /* Routines to load 1-4 rvec from 4 places.
 131  * We mainly use these to load coordinates. The extra routines
 132  * are very efficient for the water-water loops, since we e.g.
 133  * know that a TIP4p water has 4 atoms, so we should load 12 floats+shuffle.
 134  */
 135
 136
 137 static gmx_inline void
 138 gmx_mm_load_shift_and_1rvec_broadcast_ps(const float * gmx_restrict xyz_shift,
 139                                          const float * gmx_restrict xyz,
 140                                          __m128 * gmx_restrict      x1,
 141                                          __m128 * gmx_restrict      y1,
 142                                          __m128 * gmx_restrict      z1)
 143 {
 144     __m128 t1, t2, t3, t4;
 145
 146     t1   = _mm_loadl_pi(_mm_setzero_ps(), (__m64 *)xyz_shift);
 147     t2   = _mm_loadl_pi(_mm_setzero_ps(), (__m64 *)xyz);
 148     t3   = _mm_load_ss(xyz_shift+2);
 149     t4   = _mm_load_ss(xyz+2);
 150     t1   = _mm_add_ps(t1, t2);
 151     t3   = _mm_add_ss(t3, t4);
 152
 153     *x1  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(0, 0, 0, 0));
 154     *y1  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(1, 1, 1, 1));
 155     *z1  = _mm_shuffle_ps(t3, t3, _MM_SHUFFLE(0, 0, 0, 0));
 156 }
 157
 158
 159 static gmx_inline void
 160 gmx_mm_load_shift_and_3rvec_broadcast_ps(const float * gmx_restrict xyz_shift,
 161                                          const float * gmx_restrict xyz,
 162                                          __m128 * gmx_restrict x1, __m128 * gmx_restrict y1, __m128 * gmx_restrict z1,
 163                                          __m128 * gmx_restrict x2, __m128 * gmx_restrict y2, __m128 * gmx_restrict z2,
 164                                          __m128 * gmx_restrict x3, __m128 * gmx_restrict y3, __m128 * gmx_restrict z3)
 165 {
 166     __m128 tA, tB;
 167     __m128 t1, t2, t3, t4, t5, t6;
 168
 169     tA   = _mm_loadl_pi(_mm_setzero_ps(), (__m64 *)xyz_shift);
 170     tB   = _mm_load_ss(xyz_shift+2);
 171
 172     t1   = _mm_loadu_ps(xyz);
 173     t2   = _mm_loadu_ps(xyz+4);
 174     t3   = _mm_load_ss(xyz+8);
 175
 176     tA   = _mm_movelh_ps(tA, tB);
 177     t4   = _mm_shuffle_ps(tA, tA, _MM_SHUFFLE(0, 2, 1, 0));
 178     t5   = _mm_shuffle_ps(tA, tA, _MM_SHUFFLE(1, 0, 2, 1));
 179     t6   = _mm_shuffle_ps(tA, tA, _MM_SHUFFLE(2, 1, 0, 2));
 180
 181     t1   = _mm_add_ps(t1, t4);
 182     t2   = _mm_add_ps(t2, t5);
 183     t3   = _mm_add_ss(t3, t6);
 184
 185     *x1  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(0, 0, 0, 0));
 186     *y1  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(1, 1, 1, 1));
 187     *z1  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(2, 2, 2, 2));
 188     *x2  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(3, 3, 3, 3));
 189     *y2  = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(0, 0, 0, 0));
 190     *z2  = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(1, 1, 1, 1));
 191     *x3  = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(2, 2, 2, 2));
 192     *y3  = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(3, 3, 3, 3));
 193     *z3  = _mm_shuffle_ps(t3, t3, _MM_SHUFFLE(0, 0, 0, 0));
 194 }
 195
 196
 197 static gmx_inline void
 198 gmx_mm_load_shift_and_4rvec_broadcast_ps(const float * gmx_restrict xyz_shift,
 199                                          const float * gmx_restrict xyz,
 200                                          __m128 * gmx_restrict x1, __m128 * gmx_restrict y1, __m128 * gmx_restrict z1,
 201                                          __m128 * gmx_restrict x2, __m128 * gmx_restrict y2, __m128 * gmx_restrict z2,
 202                                          __m128 * gmx_restrict x3, __m128 * gmx_restrict y3, __m128 * gmx_restrict z3,
 203                                          __m128 * gmx_restrict x4, __m128 * gmx_restrict y4, __m128 * gmx_restrict z4)
 204 {
 205     __m128 tA, tB;
 206     __m128 t1, t2, t3, t4, t5, t6;
 207
 208     tA   = _mm_castpd_ps(_mm_load_sd((const double *)xyz_shift));
 209     tB   = _mm_load_ss(xyz_shift+2);
 210
 211     t1   = _mm_loadu_ps(xyz);
 212     t2   = _mm_loadu_ps(xyz+4);
 213     t3   = _mm_loadu_ps(xyz+8);
 214
 215     tA   = _mm_movelh_ps(tA, tB);
 216     t4   = _mm_shuffle_ps(tA, tA, _MM_SHUFFLE(0, 2, 1, 0));
 217     t5   = _mm_shuffle_ps(tA, tA, _MM_SHUFFLE(1, 0, 2, 1));
 218     t6   = _mm_shuffle_ps(tA, tA, _MM_SHUFFLE(2, 1, 0, 2));
 219
 220     t1   = _mm_add_ps(t1, t4);
 221     t2   = _mm_add_ps(t2, t5);
 222     t3   = _mm_add_ps(t3, t6);
 223
 224     *x1  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(0, 0, 0, 0));
 225     *y1  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(1, 1, 1, 1));
 226     *z1  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(2, 2, 2, 2));
 227     *x2  = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(3, 3, 3, 3));
 228     *y2  = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(0, 0, 0, 0));
 229     *z2  = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(1, 1, 1, 1));
 230     *x3  = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(2, 2, 2, 2));
 231     *y3  = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(3, 3, 3, 3));
 232     *z3  = _mm_shuffle_ps(t3, t3, _MM_SHUFFLE(0, 0, 0, 0));
 233     *x4  = _mm_shuffle_ps(t3, t3, _MM_SHUFFLE(1, 1, 1, 1));
 234     *y4  = _mm_shuffle_ps(t3, t3, _MM_SHUFFLE(2, 2, 2, 2));
 235     *z4  = _mm_shuffle_ps(t3, t3, _MM_SHUFFLE(3, 3, 3, 3));
 236 }
 237
 238
 239 static void
 240 gmx_mm_load_1rvec_4ptr_swizzle_ps(const float * gmx_restrict ptrA,
 241                                   const float * gmx_restrict ptrB,
 242                                   const float * gmx_restrict ptrC,
 243                                   const float * gmx_restrict ptrD,
 244                                   __m128 *      gmx_restrict x1,
 245                                   __m128 *      gmx_restrict y1,
 246                                   __m128 *      gmx_restrict z1)
 247 {
 248     __m128 t1, t2, t3, t4, t5, t6, t7, t8;
 249     t1   = _mm_castpd_ps(_mm_load_sd((const double *)ptrA));
 250     t2   = _mm_castpd_ps(_mm_load_sd((const double *)ptrB));
 251     t3   = _mm_castpd_ps(_mm_load_sd((const double *)ptrC));
 252     t4   = _mm_castpd_ps(_mm_load_sd((const double *)ptrD));
 253     t5   = _mm_load_ss(ptrA+2);
 254     t6   = _mm_load_ss(ptrB+2);
 255     t7   = _mm_load_ss(ptrC+2);
 256     t8   = _mm_load_ss(ptrD+2);
 257     t1   = _mm_unpacklo_ps(t1, t2);
 258     t3   = _mm_unpacklo_ps(t3, t4);
 259     *x1  = _mm_movelh_ps(t1, t3);
 260     *y1  = _mm_movehl_ps(t3, t1);
 261     t5   = _mm_unpacklo_ps(t5, t6);
 262     t7   = _mm_unpacklo_ps(t7, t8);
 263     *z1  = _mm_movelh_ps(t5, t7);
 264 }
 265
 266
 267 static void
 268 gmx_mm_load_3rvec_4ptr_swizzle_ps(const float * gmx_restrict ptrA,
 269                                   const float * gmx_restrict ptrB,
 270                                   const float * gmx_restrict ptrC,
 271                                   const float * gmx_restrict ptrD,
 272                                   __m128 * gmx_restrict x1, __m128 * gmx_restrict y1, __m128 * gmx_restrict z1,
 273                                   __m128 * gmx_restrict x2, __m128 * gmx_restrict y2, __m128 * gmx_restrict z2,
 274                                   __m128 * gmx_restrict x3, __m128 * gmx_restrict y3, __m128 * gmx_restrict z3)
 275 {
 276     __m128 t1, t2, t3, t4;
 277     t1            = _mm_loadu_ps(ptrA);
 278     t2            = _mm_loadu_ps(ptrB);
 279     t3            = _mm_loadu_ps(ptrC);
 280     t4            = _mm_loadu_ps(ptrD);
 281     _MM_TRANSPOSE4_PS(t1, t2, t3, t4);
 282     *x1           = t1;
 283     *y1           = t2;
 284     *z1           = t3;
 285     *x2           = t4;
 286     t1            = _mm_loadu_ps(ptrA+4);
 287     t2            = _mm_loadu_ps(ptrB+4);
 288     t3            = _mm_loadu_ps(ptrC+4);
 289     t4            = _mm_loadu_ps(ptrD+4);
 290     _MM_TRANSPOSE4_PS(t1, t2, t3, t4);
 291     *y2           = t1;
 292     *z2           = t2;
 293     *x3           = t3;
 294     *y3           = t4;
 295     t1            = _mm_load_ss(ptrA+8);
 296     t2            = _mm_load_ss(ptrB+8);
 297     t3            = _mm_load_ss(ptrC+8);
 298     t4            = _mm_load_ss(ptrD+8);
 299     t1            = _mm_unpacklo_ps(t1, t3);
 300     t3            = _mm_unpacklo_ps(t2, t4);
 301     *z3           = _mm_unpacklo_ps(t1, t3);
 302 }
 303
 304
 305 static void
 306 gmx_mm_load_4rvec_4ptr_swizzle_ps(const float * gmx_restrict ptrA,
 307                                   const float * gmx_restrict ptrB,
 308                                   const float * gmx_restrict ptrC,
 309                                   const float * gmx_restrict ptrD,
 310                                   __m128 * gmx_restrict x1, __m128 * gmx_restrict y1, __m128 * gmx_restrict z1,
 311                                   __m128 * gmx_restrict x2, __m128 * gmx_restrict y2, __m128 * gmx_restrict z2,
 312                                   __m128 * gmx_restrict x3, __m128 * gmx_restrict y3, __m128 * gmx_restrict z3,
 313                                   __m128 * gmx_restrict x4, __m128 * gmx_restrict y4, __m128 * gmx_restrict z4)
 314 {
 315     __m128 t1, t2, t3, t4;
 316     t1            = _mm_loadu_ps(ptrA);
 317     t2            = _mm_loadu_ps(ptrB);
 318     t3            = _mm_loadu_ps(ptrC);
 319     t4            = _mm_loadu_ps(ptrD);
 320     _MM_TRANSPOSE4_PS(t1, t2, t3, t4);
 321     *x1           = t1;
 322     *y1           = t2;
 323     *z1           = t3;
 324     *x2           = t4;
 325     t1            = _mm_loadu_ps(ptrA+4);
 326     t2            = _mm_loadu_ps(ptrB+4);
 327     t3            = _mm_loadu_ps(ptrC+4);
 328     t4            = _mm_loadu_ps(ptrD+4);
 329     _MM_TRANSPOSE4_PS(t1, t2, t3, t4);
 330     *y2           = t1;
 331     *z2           = t2;
 332     *x3           = t3;
 333     *y3           = t4;
 334     t1            = _mm_loadu_ps(ptrA+8);
 335     t2            = _mm_loadu_ps(ptrB+8);
 336     t3            = _mm_loadu_ps(ptrC+8);
 337     t4            = _mm_loadu_ps(ptrD+8);
 338     _MM_TRANSPOSE4_PS(t1, t2, t3, t4);
 339     *z3           = t1;
 340     *x4           = t2;
 341     *y4           = t3;
 342     *z4           = t4;
 343 }
 344
 345
 346 static void
 347 gmx_mm_decrement_1rvec_4ptr_swizzle_ps(float * gmx_restrict ptrA,
 348                                        float * gmx_restrict ptrB,
 349                                        float * gmx_restrict ptrC,
 350                                        float * gmx_restrict ptrD,
 351                                        __m128 x1, __m128 y1, __m128 z1)
 352 {
 353     __m128 t1, t2, t3, t4, t5, t6, t7, t8, t9, t10, t11, t12;
 354     t5          = _mm_unpacklo_ps(y1, z1);
 355     t6          = _mm_unpackhi_ps(y1, z1);
 356     t7          = _mm_shuffle_ps(x1, t5, _MM_SHUFFLE(1, 0, 0, 0));
 357     t8          = _mm_shuffle_ps(x1, t5, _MM_SHUFFLE(3, 2, 0, 1));
 358     t9          = _mm_shuffle_ps(x1, t6, _MM_SHUFFLE(1, 0, 0, 2));
 359     t10         = _mm_shuffle_ps(x1, t6, _MM_SHUFFLE(3, 2, 0, 3));
 360     t1          = _mm_load_ss(ptrA);
 361     t1          = _mm_loadh_pi(t1, (__m64 *)(ptrA+1));
 362     t1          = _mm_sub_ps(t1, t7);
 363     _mm_store_ss(ptrA, t1);
 364     _mm_storeh_pi((__m64 *)(ptrA+1), t1);
 365     t2          = _mm_load_ss(ptrB);
 366     t2          = _mm_loadh_pi(t2, (__m64 *)(ptrB+1));
 367     t2          = _mm_sub_ps(t2, t8);
 368     _mm_store_ss(ptrB, t2);
 369     _mm_storeh_pi((__m64 *)(ptrB+1), t2);
 370     t3          = _mm_load_ss(ptrC);
 371     t3          = _mm_loadh_pi(t3, (__m64 *)(ptrC+1));
 372     t3          = _mm_sub_ps(t3, t9);
 373     _mm_store_ss(ptrC, t3);
 374     _mm_storeh_pi((__m64 *)(ptrC+1), t3);
 375     t4          = _mm_load_ss(ptrD);
 376     t4          = _mm_loadh_pi(t4, (__m64 *)(ptrD+1));
 377     t4          = _mm_sub_ps(t4, t10);
 378     _mm_store_ss(ptrD, t4);
 379     _mm_storeh_pi((__m64 *)(ptrD+1), t4);
 380 }
 381
 382
 383
 384 #if defined (_MSC_VER) && defined(_M_IX86)
 385 /* Macro work-around since 32-bit MSVC cannot handle >3 xmm/ymm parameters */
 386 #define gmx_mm_decrement_3rvec_4ptr_swizzle_ps(ptrA, ptrB, ptrC, ptrD, \
 387                                                _x1, _y1, _z1, _x2, _y2, _z2, _x3, _y3, _z3) \
 388     { \
 389         __m128 _t1, _t2, _t3, _t4, _t5, _t6, _t7, _t8, _t9, _t10; \
 390         __m128 _t11, _t12, _t13, _t14, _t15, _t16, _t17, _t18, _t19; \
 391         __m128 _t20, _t21, _t22, _t23, _t24, _t25; \
 392         _t13         = _mm_unpackhi_ps(_x1, _y1); \
 393         _x1          = _mm_unpacklo_ps(_x1, _y1); \
 394         _t14         = _mm_unpackhi_ps(_z1, _x2); \
 395         _z1          = _mm_unpacklo_ps(_z1, _x2); \
 396         _t15         = _mm_unpackhi_ps(_y2, _z2); \
 397         _y2          = _mm_unpacklo_ps(_y2, _z2); \
 398         _t16         = _mm_unpackhi_ps(_x3, _y3); \
 399         _x3          = _mm_unpacklo_ps(_x3, _y3); \
 400         _t17         = _mm_shuffle_ps(_z3, _z3, _MM_SHUFFLE(0, 0, 0, 1)); \
 401         _t18         = _mm_movehl_ps(_z3, _z3); \
 402         _t19         = _mm_shuffle_ps(_t18, _t18, _MM_SHUFFLE(0, 0, 0, 1)); \
 403         _t20         = _mm_movelh_ps(_x1, _z1); \
 404         _t21         = _mm_movehl_ps(_z1, _x1); \
 405         _t22         = _mm_movelh_ps(_t13, _t14); \
 406         _t14         = _mm_movehl_ps(_t14, _t13); \
 407         _t23         = _mm_movelh_ps(_y2, _x3); \
 408         _t24         = _mm_movehl_ps(_x3, _y2); \
 409         _t25         = _mm_movelh_ps(_t15, _t16); \
 410         _t16         = _mm_movehl_ps(_t16, _t15); \
 411         _t1          = _mm_loadu_ps(ptrA); \
 412         _t2          = _mm_loadu_ps(ptrA+4); \
 413         _t3          = _mm_load_ss(ptrA+8); \
 414         _t1          = _mm_sub_ps(_t1, _t20); \
 415         _t2          = _mm_sub_ps(_t2, _t23); \
 416         _t3          = _mm_sub_ss(_t3, _z3); \
 417         _mm_storeu_ps(ptrA, _t1); \
 418         _mm_storeu_ps(ptrA+4, _t2); \
 419         _mm_store_ss(ptrA+8, _t3); \
 420         _t4          = _mm_loadu_ps(ptrB); \
 421         _t5          = _mm_loadu_ps(ptrB+4); \
 422         _t6          = _mm_load_ss(ptrB+8); \
 423         _t4          = _mm_sub_ps(_t4, _t21); \
 424         _t5          = _mm_sub_ps(_t5, _t24); \
 425         _t6          = _mm_sub_ss(_t6, _t17); \
 426         _mm_storeu_ps(ptrB, _t4); \
 427         _mm_storeu_ps(ptrB+4, _t5); \
 428         _mm_store_ss(ptrB+8, _t6); \
 429         _t7          = _mm_loadu_ps(ptrC); \
 430         _t8          = _mm_loadu_ps(ptrC+4); \
 431         _t9          = _mm_load_ss(ptrC+8); \
 432         _t7          = _mm_sub_ps(_t7, _t22); \
 433         _t8          = _mm_sub_ps(_t8, _t25); \
 434         _t9          = _mm_sub_ss(_t9, _t18); \
 435         _mm_storeu_ps(ptrC, _t7); \
 436         _mm_storeu_ps(ptrC+4, _t8); \
 437         _mm_store_ss(ptrC+8, _t9); \
 438         _t10         = _mm_loadu_ps(ptrD); \
 439         _t11         = _mm_loadu_ps(ptrD+4); \
 440         _t12         = _mm_load_ss(ptrD+8); \
 441         _t10         = _mm_sub_ps(_t10, _t14); \
 442         _t11         = _mm_sub_ps(_t11, _t16); \
 443         _t12         = _mm_sub_ss(_t12, _t19); \
 444         _mm_storeu_ps(ptrD, _t10); \
 445         _mm_storeu_ps(ptrD+4, _t11); \
 446         _mm_store_ss(ptrD+8, _t12); \
 447     }
 448 #else
 449 /* Real function for sane compilers */
 450 static void
 451 gmx_mm_decrement_3rvec_4ptr_swizzle_ps(float * gmx_restrict ptrA, float * gmx_restrict ptrB,
 452                                        float * gmx_restrict ptrC, float * gmx_restrict ptrD,
 453                                        __m128 x1, __m128 y1, __m128 z1,
 454                                        __m128 x2, __m128 y2, __m128 z2,
 455                                        __m128 x3, __m128 y3, __m128 z3)
 456 {
 457     __m128 t1, t2, t3, t4, t5, t6, t7, t8, t9, t10;
 458     __m128 t11, t12, t13, t14, t15, t16, t17, t18, t19;
 459     __m128 t20, t21, t22, t23, t24, t25;
 460
 461     t13         = _mm_unpackhi_ps(x1, y1);
 462     x1          = _mm_unpacklo_ps(x1, y1);
 463     t14         = _mm_unpackhi_ps(z1, x2);
 464     z1          = _mm_unpacklo_ps(z1, x2);
 465     t15         = _mm_unpackhi_ps(y2, z2);
 466     y2          = _mm_unpacklo_ps(y2, z2);
 467     t16         = _mm_unpackhi_ps(x3, y3);
 468     x3          = _mm_unpacklo_ps(x3, y3);
 469     t17         = _mm_shuffle_ps(z3, z3, _MM_SHUFFLE(0, 0, 0, 1));
 470     t18         = _mm_movehl_ps(z3, z3);
 471     t19         = _mm_shuffle_ps(t18, t18, _MM_SHUFFLE(0, 0, 0, 1));
 472     t20         = _mm_movelh_ps(x1, z1);
 473     t21         = _mm_movehl_ps(z1, x1);
 474     t22         = _mm_movelh_ps(t13, t14);
 475     t14         = _mm_movehl_ps(t14, t13);
 476     t23         = _mm_movelh_ps(y2, x3);
 477     t24         = _mm_movehl_ps(x3, y2);
 478     t25         = _mm_movelh_ps(t15, t16);
 479     t16         = _mm_movehl_ps(t16, t15);
 480     t1          = _mm_loadu_ps(ptrA);
 481     t2          = _mm_loadu_ps(ptrA+4);
 482     t3          = _mm_load_ss(ptrA+8);
 483     t1          = _mm_sub_ps(t1, t20);
 484     t2          = _mm_sub_ps(t2, t23);
 485     t3          = _mm_sub_ss(t3, z3);
 486     _mm_storeu_ps(ptrA, t1);
 487     _mm_storeu_ps(ptrA+4, t2);
 488     _mm_store_ss(ptrA+8, t3);
 489     t4          = _mm_loadu_ps(ptrB);
 490     t5          = _mm_loadu_ps(ptrB+4);
 491     t6          = _mm_load_ss(ptrB+8);
 492     t4          = _mm_sub_ps(t4, t21);
 493     t5          = _mm_sub_ps(t5, t24);
 494     t6          = _mm_sub_ss(t6, t17);
 495     _mm_storeu_ps(ptrB, t4);
 496     _mm_storeu_ps(ptrB+4, t5);
 497     _mm_store_ss(ptrB+8, t6);
 498     t7          = _mm_loadu_ps(ptrC);
 499     t8          = _mm_loadu_ps(ptrC+4);
 500     t9          = _mm_load_ss(ptrC+8);
 501     t7          = _mm_sub_ps(t7, t22);
 502     t8          = _mm_sub_ps(t8, t25);
 503     t9          = _mm_sub_ss(t9, t18);
 504     _mm_storeu_ps(ptrC, t7);
 505     _mm_storeu_ps(ptrC+4, t8);
 506     _mm_store_ss(ptrC+8, t9);
 507     t10         = _mm_loadu_ps(ptrD);
 508     t11         = _mm_loadu_ps(ptrD+4);
 509     t12         = _mm_load_ss(ptrD+8);
 510     t10         = _mm_sub_ps(t10, t14);
 511     t11         = _mm_sub_ps(t11, t16);
 512     t12         = _mm_sub_ss(t12, t19);
 513     _mm_storeu_ps(ptrD, t10);
 514     _mm_storeu_ps(ptrD+4, t11);
 515     _mm_store_ss(ptrD+8, t12);
 516 }
 517 #endif
 518
 519
 520 #if defined (_MSC_VER) && defined(_M_IX86)
 521 /* Macro work-around since 32-bit MSVC cannot handle >3 xmm/ymm parameters */
 522 #define gmx_mm_decrement_4rvec_4ptr_swizzle_ps(ptrA, ptrB, ptrC, ptrD, \
 523                                                _x1, _y1, _z1, _x2, _y2, _z2, _x3, _y3, _z3, _x4, _y4, _z4) \
 524     { \
 525         __m128 _t1, _t2, _t3, _t4, _t5, _t6, _t7, _t8, _t9, _t10, _t11; \
 526         __m128 _t12, _t13, _t14, _t15, _t16, _t17, _t18, _t19, _t20, _t21, _t22; \
 527         __m128 _t23, _t24; \
 528         _t13         = _mm_unpackhi_ps(_x1, _y1); \
 529         _x1          = _mm_unpacklo_ps(_x1, _y1); \
 530         _t14         = _mm_unpackhi_ps(_z1, _x2); \
 531         _z1          = _mm_unpacklo_ps(_z1, _x2); \
 532         _t15         = _mm_unpackhi_ps(_y2, _z2); \
 533         _y2          = _mm_unpacklo_ps(_y2, _z2); \
 534         _t16         = _mm_unpackhi_ps(_x3, _y3); \
 535         _x3          = _mm_unpacklo_ps(_x3, _y3); \
 536         _t17         = _mm_unpackhi_ps(_z3, _x4); \
 537         _z3          = _mm_unpacklo_ps(_z3, _x4); \
 538         _t18         = _mm_unpackhi_ps(_y4, _z4); \
 539         _y4          = _mm_unpacklo_ps(_y4, _z4); \
 540         _t19         = _mm_movelh_ps(_x1, _z1); \
 541         _z1          = _mm_movehl_ps(_z1, _x1); \
 542         _t20         = _mm_movelh_ps(_t13, _t14); \
 543         _t14         = _mm_movehl_ps(_t14, _t13); \
 544         _t21         = _mm_movelh_ps(_y2, _x3); \
 545         _x3          = _mm_movehl_ps(_x3, _y2); \
 546         _t22         = _mm_movelh_ps(_t15, _t16); \
 547         _t16         = _mm_movehl_ps(_t16, _t15); \
 548         _t23         = _mm_movelh_ps(_z3, _y4); \
 549         _y4          = _mm_movehl_ps(_y4, _z3); \
 550         _t24         = _mm_movelh_ps(_t17, _t18); \
 551         _t18         = _mm_movehl_ps(_t18, _t17); \
 552         _t1          = _mm_loadu_ps(ptrA); \
 553         _t2          = _mm_loadu_ps(ptrA+4); \
 554         _t3          = _mm_loadu_ps(ptrA+8); \
 555         _t1          = _mm_sub_ps(_t1, _t19); \
 556         _t2          = _mm_sub_ps(_t2, _t21); \
 557         _t3          = _mm_sub_ps(_t3, _t23); \
 558         _mm_storeu_ps(ptrA, _t1); \
 559         _mm_storeu_ps(ptrA+4, _t2); \
 560         _mm_storeu_ps(ptrA+8, _t3); \
 561         _t4          = _mm_loadu_ps(ptrB); \
 562         _t5          = _mm_loadu_ps(ptrB+4); \
 563         _t6          = _mm_loadu_ps(ptrB+8); \
 564         _t4          = _mm_sub_ps(_t4, _z1); \
 565         _t5          = _mm_sub_ps(_t5, _x3); \
 566         _t6          = _mm_sub_ps(_t6, _y4); \
 567         _mm_storeu_ps(ptrB, _t4); \
 568         _mm_storeu_ps(ptrB+4, _t5); \
 569         _mm_storeu_ps(ptrB+8, _t6); \
 570         _t7          = _mm_loadu_ps(ptrC); \
 571         _t8          = _mm_loadu_ps(ptrC+4); \
 572         _t9          = _mm_loadu_ps(ptrC+8); \
 573         _t7          = _mm_sub_ps(_t7, _t20); \
 574         _t8          = _mm_sub_ps(_t8, _t22); \
 575         _t9          = _mm_sub_ps(_t9, _t24); \
 576         _mm_storeu_ps(ptrC, _t7); \
 577         _mm_storeu_ps(ptrC+4, _t8); \
 578         _mm_storeu_ps(ptrC+8, _t9); \
 579         _t10         = _mm_loadu_ps(ptrD); \
 580         _t11         = _mm_loadu_ps(ptrD+4); \
 581         _t12         = _mm_loadu_ps(ptrD+8); \
 582         _t10         = _mm_sub_ps(_t10, _t14); \
 583         _t11         = _mm_sub_ps(_t11, _t16); \
 584         _t12         = _mm_sub_ps(_t12, _t18); \
 585         _mm_storeu_ps(ptrD, _t10); \
 586         _mm_storeu_ps(ptrD+4, _t11); \
 587         _mm_storeu_ps(ptrD+8, _t12); \
 588     }
 589 #else
 590 /* Real function for sane compilers */
 591 static void
 592 gmx_mm_decrement_4rvec_4ptr_swizzle_ps(float * gmx_restrict ptrA, float * gmx_restrict ptrB,
 593                                        float * gmx_restrict ptrC, float * gmx_restrict ptrD,
 594                                        __m128 x1, __m128 y1, __m128 z1,
 595                                        __m128 x2, __m128 y2, __m128 z2,
 596                                        __m128 x3, __m128 y3, __m128 z3,
 597                                        __m128 x4, __m128 y4, __m128 z4)
 598 {
 599     __m128 t1, t2, t3, t4, t5, t6, t7, t8, t9, t10, t11;
 600     __m128 t12, t13, t14, t15, t16, t17, t18, t19, t20, t21, t22;
 601     __m128 t23, t24;
 602     t13         = _mm_unpackhi_ps(x1, y1);
 603     x1          = _mm_unpacklo_ps(x1, y1);
 604     t14         = _mm_unpackhi_ps(z1, x2);
 605     z1          = _mm_unpacklo_ps(z1, x2);
 606     t15         = _mm_unpackhi_ps(y2, z2);
 607     y2          = _mm_unpacklo_ps(y2, z2);
 608     t16         = _mm_unpackhi_ps(x3, y3);
 609     x3          = _mm_unpacklo_ps(x3, y3);
 610     t17         = _mm_unpackhi_ps(z3, x4);
 611     z3          = _mm_unpacklo_ps(z3, x4);
 612     t18         = _mm_unpackhi_ps(y4, z4);
 613     y4          = _mm_unpacklo_ps(y4, z4);
 614     t19         = _mm_movelh_ps(x1, z1);
 615     z1          = _mm_movehl_ps(z1, x1);
 616     t20         = _mm_movelh_ps(t13, t14);
 617     t14         = _mm_movehl_ps(t14, t13);
 618     t21         = _mm_movelh_ps(y2, x3);
 619     x3          = _mm_movehl_ps(x3, y2);
 620     t22         = _mm_movelh_ps(t15, t16);
 621     t16         = _mm_movehl_ps(t16, t15);
 622     t23         = _mm_movelh_ps(z3, y4);
 623     y4          = _mm_movehl_ps(y4, z3);
 624     t24         = _mm_movelh_ps(t17, t18);
 625     t18         = _mm_movehl_ps(t18, t17);
 626     t1          = _mm_loadu_ps(ptrA);
 627     t2          = _mm_loadu_ps(ptrA+4);
 628     t3          = _mm_loadu_ps(ptrA+8);
 629     t1          = _mm_sub_ps(t1, t19);
 630     t2          = _mm_sub_ps(t2, t21);
 631     t3          = _mm_sub_ps(t3, t23);
 632     _mm_storeu_ps(ptrA, t1);
 633     _mm_storeu_ps(ptrA+4, t2);
 634     _mm_storeu_ps(ptrA+8, t3);
 635     t4          = _mm_loadu_ps(ptrB);
 636     t5          = _mm_loadu_ps(ptrB+4);
 637     t6          = _mm_loadu_ps(ptrB+8);
 638     t4          = _mm_sub_ps(t4, z1);
 639     t5          = _mm_sub_ps(t5, x3);
 640     t6          = _mm_sub_ps(t6, y4);
 641     _mm_storeu_ps(ptrB, t4);
 642     _mm_storeu_ps(ptrB+4, t5);
 643     _mm_storeu_ps(ptrB+8, t6);
 644     t7          = _mm_loadu_ps(ptrC);
 645     t8          = _mm_loadu_ps(ptrC+4);
 646     t9          = _mm_loadu_ps(ptrC+8);
 647     t7          = _mm_sub_ps(t7, t20);
 648     t8          = _mm_sub_ps(t8, t22);
 649     t9          = _mm_sub_ps(t9, t24);
 650     _mm_storeu_ps(ptrC, t7);
 651     _mm_storeu_ps(ptrC+4, t8);
 652     _mm_storeu_ps(ptrC+8, t9);
 653     t10         = _mm_loadu_ps(ptrD);
 654     t11         = _mm_loadu_ps(ptrD+4);
 655     t12         = _mm_loadu_ps(ptrD+8);
 656     t10         = _mm_sub_ps(t10, t14);
 657     t11         = _mm_sub_ps(t11, t16);
 658     t12         = _mm_sub_ps(t12, t18);
 659     _mm_storeu_ps(ptrD, t10);
 660     _mm_storeu_ps(ptrD+4, t11);
 661     _mm_storeu_ps(ptrD+8, t12);
 662 }
 663 #endif
 664
 665
 666 static gmx_inline void
 667 gmx_mm_update_iforce_1atom_swizzle_ps(__m128 fix1, __m128 fiy1, __m128 fiz1,
 668                                       float * gmx_restrict fptr,
 669                                       float * gmx_restrict fshiftptr)
 670 {
 671     __m128 t1, t2, t3;
 672
 673     /* transpose data */
 674     t1 = fix1;
 675     _MM_TRANSPOSE4_PS(fix1, t1, fiy1, fiz1);
 676     fix1 = _mm_add_ps(_mm_add_ps(fix1, t1), _mm_add_ps(fiy1, fiz1));
 677
 678     t2 = _mm_load_ss(fptr);
 679     t2 = _mm_loadh_pi(t2, (__m64 *)(fptr+1));
 680     t3 = _mm_load_ss(fshiftptr);
 681     t3 = _mm_loadh_pi(t3, (__m64 *)(fshiftptr+1));
 682
 683     t2 = _mm_add_ps(t2, fix1);
 684     t3 = _mm_add_ps(t3, fix1);
 685
 686     _mm_store_ss(fptr, t2);
 687     _mm_storeh_pi((__m64 *)(fptr+1), t2);
 688     _mm_store_ss(fshiftptr, t3);
 689     _mm_storeh_pi((__m64 *)(fshiftptr+1), t3);
 690 }
 691
 692 #if defined (_MSC_VER) && defined(_M_IX86)
 693 /* Macro work-around since 32-bit MSVC cannot handle >3 xmm/ymm parameters */
 694 #define gmx_mm_update_iforce_3atom_swizzle_ps(fix1, fiy1, fiz1, fix2, fiy2, fiz2, fix3, fiy3, fiz3, \
 695                                               fptr, fshiftptr) \
 696     { \
 697         __m128 _t1, _t2, _t3, _t4; \
 698 \
 699         _MM_TRANSPOSE4_PS(fix1, fiy1, fiz1, fix2); \
 700         _MM_TRANSPOSE4_PS(fiy2, fiz2, fix3, fiy3); \
 701         _t2   = _mm_movehl_ps(_mm_setzero_ps(), fiz3); \
 702         _t1   = _mm_shuffle_ps(fiz3, fiz3, _MM_SHUFFLE(0, 0, 0, 1)); \
 703         _t3   = _mm_shuffle_ps(_t2, _t2, _MM_SHUFFLE(0, 0, 0, 1)); \
 704         fix1  = _mm_add_ps(_mm_add_ps(fix1, fiy1), _mm_add_ps(fiz1, fix2)); \
 705         fiy2  = _mm_add_ps(_mm_add_ps(fiy2, fiz2), _mm_add_ps(fix3, fiy3)); \
 706         fiz3  = _mm_add_ss(_mm_add_ps(fiz3, _t1), _mm_add_ps(_t2, _t3)); \
 707         _mm_storeu_ps(fptr,  _mm_add_ps(fix1, _mm_loadu_ps(fptr)  )); \
 708         _mm_storeu_ps(fptr+4, _mm_add_ps(fiy2, _mm_loadu_ps(fptr+4))); \
 709         _mm_store_ss (fptr+8, _mm_add_ss(fiz3, _mm_load_ss(fptr+8) )); \
 710         _t4 = _mm_load_ss(fshiftptr+2); \
 711         _t4 = _mm_loadh_pi(_t4, (__m64 *)(fshiftptr)); \
 712         _t1 = _mm_shuffle_ps(fiz3, fix1, _MM_SHUFFLE(1, 0, 0, 0)); \
 713         _t2 = _mm_shuffle_ps(fix1, fiy2, _MM_SHUFFLE(3, 2, 2, 2)); \
 714         _t3 = _mm_shuffle_ps(fiy2, fix1, _MM_SHUFFLE(3, 3, 0, 1)); \
 715         _t3 = _mm_shuffle_ps(_t3, _t3, _MM_SHUFFLE(1, 2, 0, 0)); \
 716         _t1 = _mm_add_ps(_t1, _t2); \
 717         _t3 = _mm_add_ps(_t3, _t4); \
 718         _t1 = _mm_add_ps(_t1, _t3); \
 719         _mm_store_ss(fshiftptr+2, _t1); \
 720         _mm_storeh_pi((__m64 *)(fshiftptr), _t1); \
 721     }
 722 #else
 723 /* Real function for sane compilers */
 724 static gmx_inline void
 725 gmx_mm_update_iforce_3atom_swizzle_ps(__m128 fix1, __m128 fiy1, __m128 fiz1,
 726                                       __m128 fix2, __m128 fiy2, __m128 fiz2,
 727                                       __m128 fix3, __m128 fiy3, __m128 fiz3,
 728                                       float * gmx_restrict fptr,
 729                                       float * gmx_restrict fshiftptr)
 730 {
 731     __m128 t1, t2, t3, t4;
 732
 733     /* transpose data */
 734     _MM_TRANSPOSE4_PS(fix1, fiy1, fiz1, fix2);
 735     _MM_TRANSPOSE4_PS(fiy2, fiz2, fix3, fiy3);
 736     t2   = _mm_movehl_ps(_mm_setzero_ps(), fiz3);
 737     t1   = _mm_shuffle_ps(fiz3, fiz3, _MM_SHUFFLE(0, 0, 0, 1));
 738     t3   = _mm_shuffle_ps(t2, t2, _MM_SHUFFLE(0, 0, 0, 1));
 739
 740     fix1 = _mm_add_ps(_mm_add_ps(fix1, fiy1), _mm_add_ps(fiz1, fix2));
 741     fiy2 = _mm_add_ps(_mm_add_ps(fiy2, fiz2), _mm_add_ps(fix3, fiy3));
 742     fiz3 = _mm_add_ss(_mm_add_ps(fiz3, t1), _mm_add_ps(t2, t3));
 743
 744     _mm_storeu_ps(fptr,  _mm_add_ps(fix1, _mm_loadu_ps(fptr)  ));
 745     _mm_storeu_ps(fptr+4, _mm_add_ps(fiy2, _mm_loadu_ps(fptr+4)));
 746     _mm_store_ss (fptr+8, _mm_add_ss(fiz3, _mm_load_ss(fptr+8) ));
 747
 748     t4 = _mm_load_ss(fshiftptr+2);
 749     t4 = _mm_loadh_pi(t4, (__m64 *)(fshiftptr));
 750
 751     t1 = _mm_shuffle_ps(fiz3, fix1, _MM_SHUFFLE(1, 0, 0, 0)); /* fiy1 fix1  -   fiz3 */
 752     t2 = _mm_shuffle_ps(fix1, fiy2, _MM_SHUFFLE(3, 2, 2, 2)); /* fiy3 fix3  -   fiz1 */
 753     t3 = _mm_shuffle_ps(fiy2, fix1, _MM_SHUFFLE(3, 3, 0, 1)); /* fix2 fix2 fiy2 fiz2 */
 754     t3 = _mm_shuffle_ps(t3, t3, _MM_SHUFFLE(1, 2, 0, 0));     /* fiy2 fix2  -   fiz2 */
 755
 756     t1 = _mm_add_ps(t1, t2);
 757     t3 = _mm_add_ps(t3, t4);
 758     t1 = _mm_add_ps(t1, t3); /* y x - z */
 759
 760     _mm_store_ss(fshiftptr+2, t1);
 761     _mm_storeh_pi((__m64 *)(fshiftptr), t1);
 762 }
 763 #endif
 764
 765 #if defined (_MSC_VER) && defined(_M_IX86)
 766 /* Macro work-around since 32-bit MSVC cannot handle >3 xmm/ymm parameters */
 767 #define gmx_mm_update_iforce_4atom_swizzle_ps(fix1, fiy1, fiz1, fix2, fiy2, fiz2, fix3, fiy3, fiz3, fix4, fiy4, fiz4, \
 768                                               fptr, fshiftptr) \
 769     { \
 770         __m128 _t1, _t2, _t3, _t4, _t5; \
 771         _MM_TRANSPOSE4_PS(fix1, fiy1, fiz1, fix2); \
 772         _MM_TRANSPOSE4_PS(fiy2, fiz2, fix3, fiy3); \
 773         _MM_TRANSPOSE4_PS(fiz3, fix4, fiy4, fiz4); \
 774         fix1 = _mm_add_ps(_mm_add_ps(fix1, fiy1), _mm_add_ps(fiz1, fix2)); \
 775         fiy2 = _mm_add_ps(_mm_add_ps(fiy2, fiz2), _mm_add_ps(fix3, fiy3)); \
 776         fiz3 = _mm_add_ps(_mm_add_ps(fiz3, fix4), _mm_add_ps(fiy4, fiz4)); \
 777         _mm_storeu_ps(fptr,  _mm_add_ps(fix1, _mm_loadu_ps(fptr)  )); \
 778         _mm_storeu_ps(fptr+4, _mm_add_ps(fiy2, _mm_loadu_ps(fptr+4))); \
 779         _mm_storeu_ps(fptr+8, _mm_add_ps(fiz3, _mm_loadu_ps(fptr+8))); \
 780         _t5 = _mm_load_ss(fshiftptr+2); \
 781         _t5 = _mm_loadh_pi(_t5, (__m64 *)(fshiftptr)); \
 782         _t1 = _mm_shuffle_ps(fix1, fix1, _MM_SHUFFLE(1, 0, 2, 2)); \
 783         _t2 = _mm_shuffle_ps(fiy2, fiy2, _MM_SHUFFLE(3, 2, 1, 1)); \
 784         _t3 = _mm_shuffle_ps(fiz3, fiz3, _MM_SHUFFLE(2, 1, 0, 0)); \
 785         _t4 = _mm_shuffle_ps(fix1, fiy2, _MM_SHUFFLE(0, 0, 3, 3)); \
 786         _t4 = _mm_shuffle_ps(fiz3, _t4, _MM_SHUFFLE(2, 0, 3, 3)); \
 787         _t1 = _mm_add_ps(_t1, _t2); \
 788         _t3 = _mm_add_ps(_t3, _t4); \
 789         _t1 = _mm_add_ps(_t1, _t3); \
 790         _t5 = _mm_add_ps(_t5, _t1); \
 791         _mm_store_ss(fshiftptr+2, _t5); \
 792         _mm_storeh_pi((__m64 *)(fshiftptr), _t5); \
 793     }
 794 #else
 795 /* Real function for sane compilers */
 796 static gmx_inline void
 797 gmx_mm_update_iforce_4atom_swizzle_ps(__m128 fix1, __m128 fiy1, __m128 fiz1,
 798                                       __m128 fix2, __m128 fiy2, __m128 fiz2,
 799                                       __m128 fix3, __m128 fiy3, __m128 fiz3,
 800                                       __m128 fix4, __m128 fiy4, __m128 fiz4,
 801                                       float * gmx_restrict fptr,
 802                                       float * gmx_restrict fshiftptr)
 803 {
 804     __m128 t1, t2, t3, t4, t5;
 805
 806     /* transpose data */
 807     _MM_TRANSPOSE4_PS(fix1, fiy1, fiz1, fix2);
 808     _MM_TRANSPOSE4_PS(fiy2, fiz2, fix3, fiy3);
 809     _MM_TRANSPOSE4_PS(fiz3, fix4, fiy4, fiz4);
 810
 811     fix1 = _mm_add_ps(_mm_add_ps(fix1, fiy1), _mm_add_ps(fiz1, fix2));
 812     fiy2 = _mm_add_ps(_mm_add_ps(fiy2, fiz2), _mm_add_ps(fix3, fiy3));
 813     fiz3 = _mm_add_ps(_mm_add_ps(fiz3, fix4), _mm_add_ps(fiy4, fiz4));
 814
 815     _mm_storeu_ps(fptr,  _mm_add_ps(fix1, _mm_loadu_ps(fptr)  ));
 816     _mm_storeu_ps(fptr+4, _mm_add_ps(fiy2, _mm_loadu_ps(fptr+4)));
 817     _mm_storeu_ps(fptr+8, _mm_add_ps(fiz3, _mm_loadu_ps(fptr+8)));
 818
 819     t5 = _mm_load_ss(fshiftptr+2);
 820     t5 = _mm_loadh_pi(t5, (__m64 *)(fshiftptr));
 821
 822     t1 = _mm_shuffle_ps(fix1, fix1, _MM_SHUFFLE(1, 0, 2, 2));
 823     t2 = _mm_shuffle_ps(fiy2, fiy2, _MM_SHUFFLE(3, 2, 1, 1));
 824     t3 = _mm_shuffle_ps(fiz3, fiz3, _MM_SHUFFLE(2, 1, 0, 0));
 825     t4 = _mm_shuffle_ps(fix1, fiy2, _MM_SHUFFLE(0, 0, 3, 3));
 826     t4 = _mm_shuffle_ps(fiz3, t4, _MM_SHUFFLE(2, 0, 3, 3));
 827
 828     t1 = _mm_add_ps(t1, t2);
 829     t3 = _mm_add_ps(t3, t4);
 830     t1 = _mm_add_ps(t1, t3);
 831     t5 = _mm_add_ps(t5, t1);
 832
 833     _mm_store_ss(fshiftptr+2, t5);
 834     _mm_storeh_pi((__m64 *)(fshiftptr), t5);
 835 }
 836 #endif
 837
 838
 839 static void
 840 gmx_mm_update_1pot_ps(__m128 pot1, float * gmx_restrict ptrA)
 841 {
 842     pot1 = _mm_add_ps(pot1, _mm_movehl_ps(_mm_setzero_ps(), pot1));
 843     pot1 = _mm_add_ps(pot1, _mm_shuffle_ps(pot1, pot1, _MM_SHUFFLE(0, 0, 0, 1)));
 844     _mm_store_ss(ptrA, _mm_add_ss(pot1, _mm_load_ss(ptrA)));
 845 }
 846
 847 static void
 848 gmx_mm_update_2pot_ps(__m128 pot1, float * gmx_restrict ptrA,
 849                       __m128 pot2, float * gmx_restrict ptrB)
 850 {
 851     __m128 t1, t2;
 852     t1   = _mm_movehl_ps(pot2, pot1);
 853     t2   = _mm_movelh_ps(pot1, pot2);
 854     t1   = _mm_add_ps(t1, t2);
 855     t2   = _mm_shuffle_ps(t1, t1, _MM_SHUFFLE(3, 3, 1, 1));
 856     pot1 = _mm_add_ps(t1, t2);
 857     pot2 = _mm_movehl_ps(t2, pot1);
 858     _mm_store_ss(ptrA, _mm_add_ss(pot1, _mm_load_ss(ptrA)));
 859     _mm_store_ss(ptrB, _mm_add_ss(pot2, _mm_load_ss(ptrB)));
 860 }
 861
 862
 863 #endif /* _kernelutil_x86_sse2_single_h_ */