2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2016,2017,2018,2019, 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.
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.
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.
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.
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.
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.
37 #include "printhardware.h"
46 #include "gromacs/gpu_utils/gpu_utils.h"
47 #include "gromacs/hardware/cpuinfo.h"
48 #include "gromacs/hardware/hardwaretopology.h"
49 #include "gromacs/hardware/hw_info.h"
50 #include "gromacs/hardware/identifyavx512fmaunits.h"
51 #include "gromacs/simd/support.h"
52 #include "gromacs/utility/basedefinitions.h"
53 #include "gromacs/utility/basenetwork.h"
54 #include "gromacs/utility/cstringutil.h"
55 #include "gromacs/utility/fatalerror.h"
56 #include "gromacs/utility/gmxmpi.h"
57 #include "gromacs/utility/logger.h"
58 #include "gromacs/utility/programcontext.h"
59 #include "gromacs/utility/stringutil.h"
60 #include "gromacs/utility/sysinfo.h"
62 //! Constant used to help minimize preprocessed code
63 static const bool bGPUBinary = GMX_GPU != GMX_GPU_NONE;
66 * Returns the GPU information text, one GPU per line.
68 static std::string sprint_gpus(const gmx_gpu_info_t& gpu_info)
71 std::vector<std::string> gpuStrings;
72 for (int i = 0; i < gpu_info.n_dev; i++)
74 get_gpu_device_info_string(stmp, gpu_info, i);
75 gpuStrings.push_back(gmx::formatString(" %s", stmp));
77 return gmx::joinStrings(gpuStrings, "\n");
80 /* Give a suitable fatal error or warning if the build configuration
81 and runtime CPU do not match. */
82 static void check_use_of_rdtscp_on_this_cpu(const gmx::MDLogger& mdlog, const gmx::CpuInfo& cpuInfo)
84 bool binaryUsesRdtscp = HAVE_RDTSCP;
86 const char* programName = gmx::getProgramContext().displayName();
88 if (cpuInfo.supportLevel() < gmx::CpuInfo::SupportLevel::Features)
92 GMX_LOG(mdlog.warning)
95 "The %s executable was compiled to use the rdtscp CPU instruction. "
96 "We cannot detect the features of your current CPU, but will proceed "
98 "If you get a crash, rebuild GROMACS with the GMX_USE_RDTSCP=OFF CMake "
105 bool cpuHasRdtscp = cpuInfo.feature(gmx::CpuInfo::Feature::X86_Rdtscp);
107 if (!cpuHasRdtscp && binaryUsesRdtscp)
110 "The %s executable was compiled to use the rdtscp CPU instruction. "
111 "However, this is not supported by the current hardware and continuing would "
113 "Please rebuild GROMACS with the GMX_USE_RDTSCP=OFF CMake option.",
117 if (cpuHasRdtscp && !binaryUsesRdtscp)
119 GMX_LOG(mdlog.warning)
121 .appendTextFormatted(
122 "The current CPU can measure timings more accurately than the code in\n"
123 "%s was configured to use. This might affect your simulation\n"
124 "speed as accurate timings are needed for load-balancing.\n"
125 "Please consider rebuilding %s with the GMX_USE_RDTSCP=ON CMake "
127 programName, programName);
132 static std::string detected_hardware_string(const gmx_hw_info_t* hwinfo, bool bFullCpuInfo)
136 const gmx::CpuInfo& cpuInfo = *hwinfo->cpuInfo;
137 const gmx::HardwareTopology& hwTop = *hwinfo->hardwareTopology;
139 s = gmx::formatString("\n");
140 s += gmx::formatString("Running on %d node%s with total", hwinfo->nphysicalnode,
141 hwinfo->nphysicalnode == 1 ? "" : "s");
142 if (hwinfo->ncore_tot > 0)
144 s += gmx::formatString(" %d cores,", hwinfo->ncore_tot);
146 s += gmx::formatString(" %d logical cores", hwinfo->nhwthread_tot);
147 if (hwinfo->gpu_info.bDetectGPUs)
149 s += gmx::formatString(", %d compatible GPU%s", hwinfo->ngpu_compatible_tot,
150 hwinfo->ngpu_compatible_tot == 1 ? "" : "s");
154 s += gmx::formatString(" (GPU detection deactivated)");
156 s += gmx::formatString("\n");
158 if (hwinfo->nphysicalnode > 1)
160 /* Print per node hardware feature counts */
161 if (hwinfo->ncore_max > 0)
163 s += gmx::formatString(" Cores per node: %2d", hwinfo->ncore_min);
164 if (hwinfo->ncore_max > hwinfo->ncore_min)
166 s += gmx::formatString(" - %2d", hwinfo->ncore_max);
168 s += gmx::formatString("\n");
170 s += gmx::formatString(" Logical cores per node: %2d", hwinfo->nhwthread_min);
171 if (hwinfo->nhwthread_max > hwinfo->nhwthread_min)
173 s += gmx::formatString(" - %2d", hwinfo->nhwthread_max);
175 s += gmx::formatString("\n");
178 s += gmx::formatString(" Compatible GPUs per node: %2d", hwinfo->ngpu_compatible_min);
179 if (hwinfo->ngpu_compatible_max > hwinfo->ngpu_compatible_min)
181 s += gmx::formatString(" - %2d", hwinfo->ngpu_compatible_max);
183 s += gmx::formatString("\n");
184 if (hwinfo->ngpu_compatible_tot > 0)
186 if (hwinfo->bIdenticalGPUs)
188 s += gmx::formatString(" All nodes have identical type(s) of GPUs\n");
192 /* This message will also appear with identical GPU types
193 * when at least one node has no GPU.
195 s += gmx::formatString(
196 " Different nodes have different type(s) and/or order of GPUs\n");
206 gmx_gethostname(host, STRLEN);
208 MPI_Comm_rank(MPI_COMM_WORLD, &rank);
210 // TODO Use a wrapper around MPI_Get_processor_name instead.
211 s += gmx::formatString("Hardware detected on host %s (the node of MPI rank %d):\n", host, rank);
213 s += gmx::formatString("Hardware detected:\n");
215 s += gmx::formatString(" CPU info:\n");
217 s += gmx::formatString(" Vendor: %s\n", cpuInfo.vendorString().c_str());
219 s += gmx::formatString(" Brand: %s\n", cpuInfo.brandString().c_str());
223 s += gmx::formatString(" Family: %d Model: %d Stepping: %d\n", cpuInfo.family(),
224 cpuInfo.model(), cpuInfo.stepping());
226 s += gmx::formatString(" Features:");
227 for (auto& f : cpuInfo.featureSet())
229 s += gmx::formatString(" %s", gmx::CpuInfo::featureString(f).c_str());
231 s += gmx::formatString("\n");
234 if (cpuInfo.feature(gmx::CpuInfo::Feature::X86_Avx512F))
236 int avx512fmaunits = gmx::identifyAvx512FmaUnits();
237 s += gmx::formatString(" Number of AVX-512 FMA units:");
238 if (avx512fmaunits > 0)
240 s += gmx::formatString(" %d", avx512fmaunits);
241 if (avx512fmaunits == 1)
243 s += gmx::formatString(" (AVX2 is faster w/o 2 AVX-512 FMA units)");
248 s += gmx::formatString(" Cannot run AVX-512 detection - assuming 2");
250 s += gmx::formatString("\n");
253 s += gmx::formatString(" Hardware topology: ");
254 switch (hwTop.supportLevel())
256 case gmx::HardwareTopology::SupportLevel::None: s += gmx::formatString("None\n"); break;
257 case gmx::HardwareTopology::SupportLevel::LogicalProcessorCount:
258 s += gmx::formatString("Only logical processor count\n");
260 case gmx::HardwareTopology::SupportLevel::Basic: s += gmx::formatString("Basic\n"); break;
261 case gmx::HardwareTopology::SupportLevel::Full: s += gmx::formatString("Full\n"); break;
262 case gmx::HardwareTopology::SupportLevel::FullWithDevices:
263 s += gmx::formatString("Full, with devices\n");
267 if (!hwTop.isThisSystem())
269 s += gmx::formatString(" NOTE: Hardware topology cached or synthetic, not detected.\n");
270 if (char* p = std::getenv("HWLOC_XMLFILE"))
272 s += gmx::formatString(" HWLOC_XMLFILE=%s\n", p);
278 if (hwTop.supportLevel() >= gmx::HardwareTopology::SupportLevel::Basic)
280 s += gmx::formatString(" Sockets, cores, and logical processors:\n");
282 for (auto& socket : hwTop.machine().sockets)
284 s += gmx::formatString(" Socket %2d:", socket.id);
285 for (auto& c : socket.cores)
287 s += gmx::formatString(" [");
288 for (auto& t : c.hwThreads)
290 s += gmx::formatString(" %3d", t.logicalProcessorId);
292 s += gmx::formatString("]");
294 s += gmx::formatString("\n");
297 if (hwTop.supportLevel() >= gmx::HardwareTopology::SupportLevel::Full)
299 s += gmx::formatString(" Numa nodes:\n");
300 for (auto& n : hwTop.machine().numa.nodes)
302 s += gmx::formatString(" Node %2d (%zu bytes mem):", n.id, n.memory);
303 for (auto& l : n.logicalProcessorId)
305 s += gmx::formatString(" %3d", l);
307 s += gmx::formatString("\n");
309 s += gmx::formatString(" Latency:\n ");
310 for (std::size_t j = 0; j < hwTop.machine().numa.nodes.size(); j++)
312 s += gmx::formatString(" %5zu", j);
314 s += gmx::formatString("\n");
315 for (std::size_t i = 0; i < hwTop.machine().numa.nodes.size(); i++)
317 s += gmx::formatString(" %5zu", i);
318 for (std::size_t j = 0; j < hwTop.machine().numa.nodes.size(); j++)
320 s += gmx::formatString(" %5.2f", hwTop.machine().numa.relativeLatency[i][j]);
322 s += gmx::formatString("\n");
326 s += gmx::formatString(" Caches:\n");
327 for (auto& c : hwTop.machine().caches)
329 s += gmx::formatString(
330 " L%d: %zu bytes, linesize %d bytes, assoc. %d, shared %d ways\n",
331 c.level, c.size, c.linesize, c.associativity, c.shared);
334 if (hwTop.supportLevel() >= gmx::HardwareTopology::SupportLevel::FullWithDevices)
336 s += gmx::formatString(" PCI devices:\n");
337 for (auto& d : hwTop.machine().devices)
339 s += gmx::formatString(
340 " %04x:%02x:%02x.%1x Id: %04x:%04x Class: 0x%04x Numa: %d\n", d.domain,
341 d.bus, d.dev, d.func, d.vendorId, d.deviceId, d.classId, d.numaNodeId);
346 if (bGPUBinary && hwinfo->gpu_info.n_dev > 0)
348 s += gmx::formatString(" GPU info:\n");
349 s += gmx::formatString(" Number of GPUs detected: %d\n", hwinfo->gpu_info.n_dev);
350 s += sprint_gpus(hwinfo->gpu_info) + "\n";
355 void gmx_print_detected_hardware(FILE* fplog,
356 const bool warnToStdErr,
357 const gmx::MDLogger& mdlog,
358 const gmx_hw_info_t* hwinfo)
360 const gmx::CpuInfo& cpuInfo = *hwinfo->cpuInfo;
362 if (fplog != nullptr)
364 std::string detected;
366 detected = detected_hardware_string(hwinfo, TRUE);
368 fprintf(fplog, "%s\n", detected.c_str());
371 // Do not spam stderr with all our internal information unless
372 // there was something that actually went wrong; general information
373 // belongs in the logfile.
375 /* Check the compiled SIMD instruction set against that of the node
376 * with the lowest SIMD level support (skip if SIMD detection did not work)
378 if (cpuInfo.supportLevel() >= gmx::CpuInfo::SupportLevel::Features)
380 gmx::simdCheck(static_cast<gmx::SimdType>(hwinfo->simd_suggest_min), fplog, warnToStdErr);
383 /* For RDTSCP we only check on our local node and skip the MPI reduction */
384 check_use_of_rdtscp_on_this_cpu(mdlog, cpuInfo);