src/gromacs/hardware/tests/hardwaretopology.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2015,2016,2018,2019, 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  * \brief
  37  * Tests for gmx::HardwareTopology
  38  *
  39  * \author Erik Lindahl <erik.lindahl@gmail.com>
  40  * \ingroup module_hardware
  41  */
  42 #include "gmxpre.h"
  43
  44 #include "gromacs/hardware/hardwaretopology.h"
  45
  46 #include "config.h"
  47
  48 #include <algorithm>
  49
  50 #include <gtest/gtest.h>
  51
  52 #include "gromacs/utility/stringutil.h"
  53
  54 namespace
  55 {
  56
  57 // There is no way we can compare to any reference data since that
  58 // depends on the architecture, but we can at least make sure that it
  59 // works to execute the tests and that they are self-consistent
  60
  61 // Although it is not strictly an error, for the very basic execution tests
  62 // we also report if we cannot extract the hardware topology on systems
  63 // where we expect to be able to. Since this might happen to users, we
  64 // provide a bit more information and ask them to mail us in this case.
  65
  66 TEST(HardwareTopologyTest, Execute)
  67 {
  68     // There is no way we can compare to any reference data since that
  69     // depends on the architecture, but we can at least make sure that it
  70     // works to execute the tests
  71
  72     gmx::HardwareTopology hwTop(gmx::HardwareTopology::detect());
  73
  74     // If we cannot even find the number of logical processors we want to flag it
  75     EXPECT_GT(hwTop.supportLevel(), gmx::HardwareTopology::SupportLevel::None)
  76     << "Cannot determine number of processors. " << std::endl
  77     << "GROMACS might still work, but it will likely hurt your performance." << std::endl
  78     << "Please mail gmx-developers@gromacs.org so we can try to fix it.";
  79 }
  80
  81 #if GMX_USE_HWLOC
  82 TEST(HardwareTopologyTest, HwlocExecute)
  83 {
  84 #if defined(__linux__)
  85     gmx::HardwareTopology hwTop(gmx::HardwareTopology::detect());
  86
  87     // On Linux with hwloc support we should be able to get at least basic information
  88     EXPECT_GE(hwTop.supportLevel(), gmx::HardwareTopology::SupportLevel::Basic)
  89     << "Cannot determine basic hardware topology from hwloc. GROMACS will still\n" << std::endl
  90     << "work, but it might affect your performance for large nodes." << std::endl
  91     << "Please mail gmx-developers@gromacs.org so we can try to fix it.";
  92 #endif
  93 }
  94 #endif
  95
  96 TEST(HardwareTopologyTest, ProcessorSelfconsistency)
  97 {
  98     gmx::HardwareTopology hwTop(gmx::HardwareTopology::detect());
  99
 100     if (hwTop.supportLevel() >= gmx::HardwareTopology::SupportLevel::Basic)
 101     {
 102         SCOPED_TRACE(gmx::formatString("Logical Processor count %d", hwTop.machine().logicalProcessorCount));
 103
 104         int  socketsInMachine = hwTop.machine().sockets.size();
 105         int  coresPerSocket   = hwTop.machine().sockets[0].cores.size();
 106         int  hwThreadsPerCore = hwTop.machine().sockets[0].cores[0].hwThreads.size();
 107
 108         auto logicalProcessors = hwTop.machine().logicalProcessors;
 109         for (auto logicalProcessorIt = logicalProcessors.begin();
 110              logicalProcessorIt != logicalProcessors.end();
 111              ++logicalProcessorIt)
 112         {
 113             // Check that logical processor information contains
 114             // reasonable values.
 115             SCOPED_TRACE(gmx::formatString("Socket rank in machine: %d", logicalProcessorIt->socketRankInMachine));
 116             SCOPED_TRACE(gmx::formatString("Core rank in socket:    %d", logicalProcessorIt->coreRankInSocket));
 117             SCOPED_TRACE(gmx::formatString("Hw thread rank in core: %d", logicalProcessorIt->hwThreadRankInCore));
 118             EXPECT_TRUE(logicalProcessorIt->socketRankInMachine >= 0 && logicalProcessorIt->socketRankInMachine < socketsInMachine);
 119             EXPECT_TRUE(logicalProcessorIt->coreRankInSocket >= 0 && logicalProcessorIt->coreRankInSocket < coresPerSocket);
 120             EXPECT_TRUE(logicalProcessorIt->hwThreadRankInCore >= 0 && logicalProcessorIt->hwThreadRankInCore < hwThreadsPerCore);
 121             // Check that logical processor information is distinct
 122             // for each logical processor.
 123
 124             for (auto remainingLogicalProcessorIt = logicalProcessorIt + 1;
 125                  remainingLogicalProcessorIt != logicalProcessors.end();
 126                  ++remainingLogicalProcessorIt)
 127             {
 128                 SCOPED_TRACE(gmx::formatString("Other socket rank in machine: %d", remainingLogicalProcessorIt->socketRankInMachine));
 129                 SCOPED_TRACE(gmx::formatString("Other core rank in socket:    %d", remainingLogicalProcessorIt->coreRankInSocket));
 130                 SCOPED_TRACE(gmx::formatString("Other hw thread rank in core: %d", remainingLogicalProcessorIt->hwThreadRankInCore));
 131                 EXPECT_TRUE((logicalProcessorIt->socketRankInMachine != remainingLogicalProcessorIt->socketRankInMachine) ||
 132                             (logicalProcessorIt->coreRankInSocket    != remainingLogicalProcessorIt->coreRankInSocket) ||
 133                             (logicalProcessorIt->hwThreadRankInCore  != remainingLogicalProcessorIt->hwThreadRankInCore)) <<
 134                 "This pair of logical processors have the same descriptive information, which is an error";
 135             }
 136         }
 137     }
 138 }
 139
 140 TEST(HardwareTopologyTest, NumaCacheSelfconsistency)
 141 {
 142     gmx::HardwareTopology hwTop(gmx::HardwareTopology::detect());
 143
 144     if (hwTop.supportLevel() >= gmx::HardwareTopology::SupportLevel::Full)
 145     {
 146         // Check that numa node id corresponds to rank
 147         for (std::size_t i = 0; i < hwTop.machine().numa.nodes.size(); i++)
 148         {
 149             EXPECT_EQ(hwTop.machine().numa.nodes[i].id, i);
 150         }
 151
 152         // Check that the sum of numa domains is the total processor count
 153         int processorsinNumaNudes = 0;
 154         for (auto &n : hwTop.machine().numa.nodes)
 155         {
 156             processorsinNumaNudes += n.logicalProcessorId.size();
 157         }
 158         EXPECT_EQ(processorsinNumaNudes, hwTop.machine().logicalProcessorCount);
 159
 160         // Check that every processor is in a numa domain (i.e., that they are unique)
 161         std::vector<int> v(hwTop.machine().logicalProcessorCount);
 162         for (auto &elem : v)
 163         {
 164             elem = 0;
 165         }
 166         for (auto &n : hwTop.machine().numa.nodes)
 167         {
 168             for (auto &idx : n.logicalProcessorId)
 169             {
 170                 v[idx] = 1;
 171             }
 172         }
 173         int uniqueProcessorsinNumaNudes = std::count(v.begin(), v.end(), 1);
 174         EXPECT_EQ(uniqueProcessorsinNumaNudes, hwTop.machine().logicalProcessorCount);
 175
 176         // We must have some memory in a numa node
 177         for (auto &n : hwTop.machine().numa.nodes)
 178         {
 179             EXPECT_GT(n.memory, 0);
 180         }
 181
 182         // Check latency matrix size and contents
 183         EXPECT_GT(hwTop.machine().numa.baseLatency, 0);
 184         EXPECT_GT(hwTop.machine().numa.maxRelativeLatency, 0);
 185         // Check number of rows matches # numa nodes
 186         EXPECT_EQ(hwTop.machine().numa.relativeLatency.size(), hwTop.machine().numa.nodes.size());
 187         for (auto &v2 : hwTop.machine().numa.relativeLatency)
 188         {
 189             // Check that size of each row matches # numa nodes
 190             EXPECT_EQ(v2.size(), hwTop.machine().numa.nodes.size());
 191             for (auto &latency : v2)
 192             {
 193                 // Latency values should be positive
 194                 EXPECT_GT(latency, 0);
 195             }
 196         }
 197
 198         // Check cache. The hwloc cache detection is fragile and can report
 199         // 0 for line size or associativity (=unknown), so we just check the size.
 200         for (auto &c : hwTop.machine().caches)
 201         {
 202             EXPECT_GT(c.size, 0);
 203         }
 204     }
 205 }
 206
 207
 208 } // namespace