src/gromacs/mdlib/settle_gpu_internal_sycl.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2019,2020,2021, 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 /*! \internal \file
  36  *
  37  * \brief SYCL-specific routines for the GPU implementation of SETTLE constraints algorithm.
  38  *
  39  * \author Artem Zhmurov <zhmurov@gmail.com>
  40  *
  41  * \ingroup module_mdlib
  42  */
  43
  44 #include "settle_gpu_internal.h"
  45
  46 #include "gromacs/gpu_utils/devicebuffer.h"
  47 #include "gromacs/gpu_utils/sycl_kernel_utils.h"
  48 #include "gromacs/pbcutil/pbc_aiuc_sycl.h"
  49 #include "gromacs/utility/gmxassert.h"
  50 #include "gromacs/utility/template_mp.h"
  51
  52 namespace gmx
  53 {
  54
  55 using cl::sycl::access::fence_space;
  56 using cl::sycl::access::mode;
  57 using cl::sycl::access::target;
  58
  59 //! Number of work-items in a work-group
  60 constexpr static int sc_workGroupSize = 256;
  61
  62 template<bool updateVelocities, bool computeVirial>
  63 auto settleKernel(cl::sycl::handler&                                           cgh,
  64                   const int                                                    numSettles,
  65                   DeviceAccessor<WaterMolecule, mode::read>                    a_settles,
  66                   SettleParameters                                             pars,
  67                   DeviceAccessor<Float3, mode::read>                           a_x,
  68                   DeviceAccessor<Float3, mode::read_write>                     a_xp,
  69                   float                                                        invdt,
  70                   OptionalAccessor<Float3, mode::read_write, updateVelocities> a_v,
  71                   OptionalAccessor<float, mode_atomic, computeVirial>          a_virialScaled,
  72                   PbcAiuc                                                      pbcAiuc)
  73 {
  74     cgh.require(a_settles);
  75     cgh.require(a_x);
  76     cgh.require(a_xp);
  77     if constexpr (updateVelocities)
  78     {
  79         cgh.require(a_v);
  80     }
  81     if constexpr (computeVirial)
  82     {
  83         cgh.require(a_virialScaled);
  84     }
  85
  86     // shmem buffer for i x+q pre-loading
  87     auto sm_threadVirial = [&]() {
  88         if constexpr (computeVirial)
  89         {
  90             return cl::sycl::accessor<float, 1, mode::read_write, target::local>(
  91                     cl::sycl::range<1>(sc_workGroupSize * 6), cgh);
  92         }
  93         else
  94         {
  95             return nullptr;
  96         }
  97     }();
  98
  99     return [=](cl::sycl::nd_item<1> itemIdx) {
 100         constexpr float almost_zero = real(1e-12);
 101         const int       settleIdx   = itemIdx.get_global_linear_id();
 102         const int       threadIdx = itemIdx.get_local_linear_id(); // Work-item index in work-group
 103         assert(itemIdx.get_local_range(0) == sc_workGroupSize);
 104         // These are the indexes of three atoms in a single 'water' molecule.
 105         // TODO Can be reduced to one integer if atoms are consecutive in memory.
 106         if (settleIdx < numSettles)
 107         {
 108             WaterMolecule indices = a_settles[settleIdx];
 109
 110             const Float3 x_ow1 = a_x[indices.ow1];
 111             const Float3 x_hw2 = a_x[indices.hw2];
 112             const Float3 x_hw3 = a_x[indices.hw3];
 113
 114             const Float3 xprime_ow1 = a_xp[indices.ow1];
 115             const Float3 xprime_hw2 = a_xp[indices.hw2];
 116             const Float3 xprime_hw3 = a_xp[indices.hw3];
 117
 118             Float3 dist21;
 119             pbcDxAiucSycl(pbcAiuc, x_hw2, x_ow1, dist21);
 120             Float3 dist31;
 121             pbcDxAiucSycl(pbcAiuc, x_hw3, x_ow1, dist31);
 122             Float3 doh2;
 123             pbcDxAiucSycl(pbcAiuc, xprime_hw2, xprime_ow1, doh2);
 124
 125             Float3 doh3;
 126             pbcDxAiucSycl(pbcAiuc, xprime_hw3, xprime_ow1, doh3);
 127
 128             Float3 a1 = (doh2 + doh3) * (-pars.wh);
 129
 130             Float3 b1 = doh2 + a1;
 131
 132             Float3 c1 = doh3 + a1;
 133
 134             float xakszd = dist21[YY] * dist31[ZZ] - dist21[ZZ] * dist31[YY];
 135             float yakszd = dist21[ZZ] * dist31[XX] - dist21[XX] * dist31[ZZ];
 136             float zakszd = dist21[XX] * dist31[YY] - dist21[YY] * dist31[XX];
 137
 138             float xaksxd = a1[YY] * zakszd - a1[ZZ] * yakszd;
 139             float yaksxd = a1[ZZ] * xakszd - a1[XX] * zakszd;
 140             float zaksxd = a1[XX] * yakszd - a1[YY] * xakszd;
 141
 142             float xaksyd = yakszd * zaksxd - zakszd * yaksxd;
 143             float yaksyd = zakszd * xaksxd - xakszd * zaksxd;
 144             float zaksyd = xakszd * yaksxd - yakszd * xaksxd;
 145
 146             float axlng = cl::sycl::rsqrt(xaksxd * xaksxd + yaksxd * yaksxd + zaksxd * zaksxd);
 147             float aylng = cl::sycl::rsqrt(xaksyd * xaksyd + yaksyd * yaksyd + zaksyd * zaksyd);
 148             float azlng = cl::sycl::rsqrt(xakszd * xakszd + yakszd * yakszd + zakszd * zakszd);
 149
 150             // TODO {1,2,3} indexes should be swapped with {.x, .y, .z} components.
 151             //      This way, we will be able to use vector ops more.
 152             Float3 trns1, trns2, trns3;
 153
 154             trns1[XX] = xaksxd * axlng;
 155             trns2[XX] = yaksxd * axlng;
 156             trns3[XX] = zaksxd * axlng;
 157
 158             trns1[YY] = xaksyd * aylng;
 159             trns2[YY] = yaksyd * aylng;
 160             trns3[YY] = zaksyd * aylng;
 161
 162             trns1[ZZ] = xakszd * azlng;
 163             trns2[ZZ] = yakszd * azlng;
 164             trns3[ZZ] = zakszd * azlng;
 165
 166
 167             Float2 b0d, c0d;
 168
 169             b0d[XX] = trns1[XX] * dist21[XX] + trns2[XX] * dist21[YY] + trns3[XX] * dist21[ZZ];
 170             b0d[YY] = trns1[YY] * dist21[XX] + trns2[YY] * dist21[YY] + trns3[YY] * dist21[ZZ];
 171
 172             c0d[XX] = trns1[XX] * dist31[XX] + trns2[XX] * dist31[YY] + trns3[XX] * dist31[ZZ];
 173             c0d[YY] = trns1[YY] * dist31[XX] + trns2[YY] * dist31[YY] + trns3[YY] * dist31[ZZ];
 174
 175             Float3 b1d, c1d;
 176
 177             float a1d_z = trns1[ZZ] * a1[XX] + trns2[ZZ] * a1[YY] + trns3[ZZ] * a1[ZZ];
 178
 179             b1d[XX] = trns1[XX] * b1[XX] + trns2[XX] * b1[YY] + trns3[XX] * b1[ZZ];
 180             b1d[YY] = trns1[YY] * b1[XX] + trns2[YY] * b1[YY] + trns3[YY] * b1[ZZ];
 181             b1d[ZZ] = trns1[ZZ] * b1[XX] + trns2[ZZ] * b1[YY] + trns3[ZZ] * b1[ZZ];
 182
 183             c1d[XX] = trns1[XX] * c1[XX] + trns2[XX] * c1[YY] + trns3[XX] * c1[ZZ];
 184             c1d[YY] = trns1[YY] * c1[XX] + trns2[YY] * c1[YY] + trns3[YY] * c1[ZZ];
 185             c1d[ZZ] = trns1[ZZ] * c1[XX] + trns2[ZZ] * c1[YY] + trns3[ZZ] * c1[ZZ];
 186
 187
 188             const float sinphi = a1d_z * cl::sycl::rsqrt(pars.ra * pars.ra);
 189             float       tmp2   = 1.0F - sinphi * sinphi;
 190
 191             if (almost_zero > tmp2)
 192             {
 193                 tmp2 = almost_zero;
 194             }
 195
 196             const float tmp    = cl::sycl::rsqrt(tmp2);
 197             const float cosphi = tmp2 * tmp;
 198             const float sinpsi = (b1d[ZZ] - c1d[ZZ]) * pars.irc2 * tmp;
 199             tmp2               = 1.0F - sinpsi * sinpsi;
 200
 201             const float cospsi = tmp2 * cl::sycl::rsqrt(tmp2);
 202
 203             const float a2d_y = pars.ra * cosphi;
 204             const float b2d_x = -pars.rc * cospsi;
 205             const float t1    = -pars.rb * cosphi;
 206             const float t2    = pars.rc * sinpsi * sinphi;
 207             const float b2d_y = t1 - t2;
 208             const float c2d_y = t1 + t2;
 209
 210             /*     --- Step3  al,be,ga            --- */
 211             const float alpha = b2d_x * (b0d[XX] - c0d[XX]) + b0d[YY] * b2d_y + c0d[YY] * c2d_y;
 212             const float beta  = b2d_x * (c0d[YY] - b0d[YY]) + b0d[XX] * b2d_y + c0d[XX] * c2d_y;
 213             const float gamma =
 214                     b0d[XX] * b1d[YY] - b1d[XX] * b0d[YY] + c0d[XX] * c1d[YY] - c1d[XX] * c0d[YY];
 215             const float al2be2 = alpha * alpha + beta * beta;
 216             tmp2               = (al2be2 - gamma * gamma);
 217             const float sinthe = (alpha * gamma - beta * tmp2 * cl::sycl::rsqrt(tmp2))
 218                                  * cl::sycl::rsqrt(al2be2 * al2be2);
 219
 220             /*  --- Step4  A3' --- */
 221             tmp2         = 1.0F - sinthe * sinthe;
 222             float costhe = tmp2 * cl::sycl::rsqrt(tmp2);
 223
 224             Float3 a3d, b3d, c3d;
 225
 226             a3d[XX] = -a2d_y * sinthe;
 227             a3d[YY] = a2d_y * costhe;
 228             a3d[ZZ] = a1d_z;
 229             b3d[XX] = b2d_x * costhe - b2d_y * sinthe;
 230             b3d[YY] = b2d_x * sinthe + b2d_y * costhe;
 231             b3d[ZZ] = b1d[ZZ];
 232             c3d[XX] = -b2d_x * costhe - c2d_y * sinthe;
 233             c3d[YY] = -b2d_x * sinthe + c2d_y * costhe;
 234             c3d[ZZ] = c1d[ZZ];
 235
 236             /*    --- Step5  A3 --- */
 237             Float3 a3, b3, c3;
 238
 239             a3[XX] = trns1[XX] * a3d[XX] + trns1[YY] * a3d[YY] + trns1[ZZ] * a3d[ZZ];
 240             a3[YY] = trns2[XX] * a3d[XX] + trns2[YY] * a3d[YY] + trns2[ZZ] * a3d[ZZ];
 241             a3[ZZ] = trns3[XX] * a3d[XX] + trns3[YY] * a3d[YY] + trns3[ZZ] * a3d[ZZ];
 242
 243             b3[XX] = trns1[XX] * b3d[XX] + trns1[YY] * b3d[YY] + trns1[ZZ] * b3d[ZZ];
 244             b3[YY] = trns2[XX] * b3d[XX] + trns2[YY] * b3d[YY] + trns2[ZZ] * b3d[ZZ];
 245             b3[ZZ] = trns3[XX] * b3d[XX] + trns3[YY] * b3d[YY] + trns3[ZZ] * b3d[ZZ];
 246
 247             c3[XX] = trns1[XX] * c3d[XX] + trns1[YY] * c3d[YY] + trns1[ZZ] * c3d[ZZ];
 248             c3[YY] = trns2[XX] * c3d[XX] + trns2[YY] * c3d[YY] + trns2[ZZ] * c3d[ZZ];
 249             c3[ZZ] = trns3[XX] * c3d[XX] + trns3[YY] * c3d[YY] + trns3[ZZ] * c3d[ZZ];
 250
 251
 252             /* Compute and store the corrected new coordinate */
 253             const Float3 dxOw1 = a3 - a1;
 254             const Float3 dxHw2 = b3 - b1;
 255             const Float3 dxHw3 = c3 - c1;
 256
 257             a_xp[indices.ow1] = xprime_ow1 + dxOw1;
 258             a_xp[indices.hw2] = xprime_hw2 + dxHw2;
 259             a_xp[indices.hw3] = xprime_hw3 + dxHw3;
 260
 261             if constexpr (updateVelocities)
 262             {
 263                 Float3 v_ow1 = a_v[indices.ow1];
 264                 Float3 v_hw2 = a_v[indices.hw2];
 265                 Float3 v_hw3 = a_v[indices.hw3];
 266
 267                 /* Add the position correction divided by dt to the velocity */
 268                 v_ow1 = dxOw1 * invdt + v_ow1;
 269                 v_hw2 = dxHw2 * invdt + v_hw2;
 270                 v_hw3 = dxHw3 * invdt + v_hw3;
 271
 272                 a_v[indices.ow1] = v_ow1;
 273                 a_v[indices.hw2] = v_hw2;
 274                 a_v[indices.hw3] = v_hw3;
 275             }
 276
 277             if constexpr (computeVirial)
 278             {
 279                 Float3 mdb = pars.mH * dxHw2;
 280                 Float3 mdc = pars.mH * dxHw3;
 281                 Float3 mdo = pars.mO * dxOw1 + mdb + mdc;
 282
 283                 sm_threadVirial[0 * sc_workGroupSize + threadIdx] =
 284                         -(x_ow1[0] * mdo[0] + dist21[0] * mdb[0] + dist31[0] * mdc[0]);
 285                 sm_threadVirial[1 * sc_workGroupSize + threadIdx] =
 286                         -(x_ow1[0] * mdo[1] + dist21[0] * mdb[1] + dist31[0] * mdc[1]);
 287                 sm_threadVirial[2 * sc_workGroupSize + threadIdx] =
 288                         -(x_ow1[0] * mdo[2] + dist21[0] * mdb[2] + dist31[0] * mdc[2]);
 289                 sm_threadVirial[3 * sc_workGroupSize + threadIdx] =
 290                         -(x_ow1[1] * mdo[1] + dist21[1] * mdb[1] + dist31[1] * mdc[1]);
 291                 sm_threadVirial[4 * sc_workGroupSize + threadIdx] =
 292                         -(x_ow1[1] * mdo[2] + dist21[1] * mdb[2] + dist31[1] * mdc[2]);
 293                 sm_threadVirial[5 * sc_workGroupSize + threadIdx] =
 294                         -(x_ow1[2] * mdo[2] + dist21[2] * mdb[2] + dist31[2] * mdc[2]);
 295             }
 296         }
 297         else // settleIdx < numSettles
 298         {
 299             // Filling data for dummy threads with zeroes
 300             if constexpr (computeVirial)
 301             {
 302                 for (int d = 0; d < 6; d++)
 303                 {
 304                     sm_threadVirial[d * sc_workGroupSize + threadIdx] = 0.0F;
 305                 }
 306             }
 307         }
 308
 309         // Basic reduction for the values inside single thread block
 310         // TODO what follows should be separated out as a standard virial reduction subroutine
 311         if constexpr (computeVirial)
 312         {
 313             // This is to ensure that all threads saved the data before reduction starts
 314             subGroupBarrier(itemIdx);
 315             constexpr int blockSize    = sc_workGroupSize;
 316             const int     subGroupSize = itemIdx.get_sub_group().get_max_local_range()[0];
 317             // Reduce up to one virial per thread block
 318             // All blocks are divided by half, the first half of threads sums
 319             // two virials. Then the first half is divided by two and the first half
 320             // of it sums two values... The procedure continues until only one thread left.
 321             // Only works if the threads per blocks is a power of two, hence the assertion.
 322             static_assert(gmx::isPowerOfTwo(sc_workGroupSize));
 323             for (int divideBy = 2; divideBy <= blockSize; divideBy *= 2)
 324             {
 325                 int dividedAt = blockSize / divideBy;
 326                 if (threadIdx < dividedAt)
 327                 {
 328                     for (int d = 0; d < 6; d++)
 329                     {
 330                         sm_threadVirial[d * blockSize + threadIdx] +=
 331                                 sm_threadVirial[d * blockSize + (threadIdx + dividedAt)];
 332                     }
 333                 }
 334                 if (dividedAt > subGroupSize / 2)
 335                 {
 336                     subGroupBarrier(itemIdx);
 337                 }
 338             }
 339             // First 6 threads in the block add the 6 components of virial to the global memory address
 340             if (threadIdx < 6)
 341             {
 342                 atomicFetchAdd(a_virialScaled, threadIdx, sm_threadVirial[threadIdx * blockSize]);
 343             }
 344         }
 345     };
 346 }
 347
 348 // SYCL 1.2.1 requires providing a unique type for a kernel. Should not be needed for SYCL2020.
 349 template<bool updateVelocities, bool computeVirial>
 350 class SettleKernelName;
 351
 352 template<bool updateVelocities, bool computeVirial, class... Args>
 353 static cl::sycl::event launchSettleKernel(const DeviceStream& deviceStream, int numSettles, Args&&... args)
 354 {
 355     // Should not be needed for SYCL2020.
 356     using kernelNameType = SettleKernelName<updateVelocities, computeVirial>;
 357
 358     const int numSettlesRoundedUp =
 359             static_cast<int>((numSettles + sc_workGroupSize - 1) / sc_workGroupSize) * sc_workGroupSize;
 360     const cl::sycl::nd_range<1> rangeAllSettles(numSettlesRoundedUp, sc_workGroupSize);
 361     cl::sycl::queue             q = deviceStream.stream();
 362
 363     cl::sycl::event e = q.submit([&](cl::sycl::handler& cgh) {
 364         auto kernel = settleKernel<updateVelocities, computeVirial>(
 365                 cgh, numSettles, std::forward<Args>(args)...);
 366         cgh.parallel_for<kernelNameType>(rangeAllSettles, kernel);
 367     });
 368
 369     return e;
 370 }
 371
 372 /*! \brief Select templated kernel and launch it. */
 373 template<class... Args>
 374 static inline cl::sycl::event launchSettleKernel(bool updateVelocities, bool computeVirial, Args&&... args)
 375 {
 376     return dispatchTemplatedFunction(
 377             [&](auto updateVelocities_, auto computeVirial_) {
 378                 return launchSettleKernel<updateVelocities_, computeVirial_>(std::forward<Args>(args)...);
 379             },
 380             updateVelocities,
 381             computeVirial);
 382 }
 383
 384
 385 void launchSettleGpuKernel(const int                          numSettles,
 386                            const DeviceBuffer<WaterMolecule>& d_settles,
 387                            const SettleParameters&            settleParameters,
 388                            const DeviceBuffer<Float3>&        d_x,
 389                            DeviceBuffer<Float3>               d_xp,
 390                            const bool                         updateVelocities,
 391                            DeviceBuffer<Float3>               d_v,
 392                            const real                         invdt,
 393                            const bool                         computeVirial,
 394                            DeviceBuffer<float>                virialScaled,
 395                            const PbcAiuc&                     pbcAiuc,
 396                            const DeviceStream&                deviceStream)
 397 {
 398
 399     launchSettleKernel(updateVelocities,
 400                        computeVirial,
 401                        deviceStream,
 402                        numSettles,
 403                        d_settles,
 404                        settleParameters,
 405                        d_x,
 406                        d_xp,
 407                        invdt,
 408                        d_v,
 409                        virialScaled,
 410                        pbcAiuc);
 411     return;
 412 }
 413
 414 } // namespace gmx