2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2016,2017,2018, 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/mdtypes/commrec.h"
52 #include "gromacs/simd/support.h"
53 #include "gromacs/utility/basedefinitions.h"
54 #include "gromacs/utility/basenetwork.h"
55 #include "gromacs/utility/cstringutil.h"
56 #include "gromacs/utility/fatalerror.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. */
83 check_use_of_rdtscp_on_this_cpu(const gmx::MDLogger
&mdlog
,
84 const gmx::CpuInfo
&cpuInfo
)
86 bool binaryUsesRdtscp
= HAVE_RDTSCP
;
88 const char *programName
= gmx::getProgramContext().displayName();
90 if (cpuInfo
.supportLevel() < gmx::CpuInfo::SupportLevel::Features
)
94 GMX_LOG(mdlog
.warning
).asParagraph().appendTextFormatted(
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 anyway. "
97 "If you get a crash, rebuild GROMACS with the GMX_USE_RDTSCP=OFF CMake option.",
103 bool cpuHasRdtscp
= cpuInfo
.feature(gmx::CpuInfo::Feature::X86_Rdtscp
);
105 if (!cpuHasRdtscp
&& binaryUsesRdtscp
)
107 gmx_fatal(FARGS
, "The %s executable was compiled to use the rdtscp CPU instruction. "
108 "However, this is not supported by the current hardware and continuing would lead to a crash. "
109 "Please rebuild GROMACS with the GMX_USE_RDTSCP=OFF CMake option.",
113 if (cpuHasRdtscp
&& !binaryUsesRdtscp
)
115 GMX_LOG(mdlog
.warning
).asParagraph().appendTextFormatted(
116 "The current CPU can measure timings more accurately than the code in\n"
117 "%s was configured to use. This might affect your simulation\n"
118 "speed as accurate timings are needed for load-balancing.\n"
119 "Please consider rebuilding %s with the GMX_USE_RDTSCP=ON CMake option.",
120 programName
, programName
);
125 static std::string
detected_hardware_string(const gmx_hw_info_t
*hwinfo
,
130 const gmx::CpuInfo
&cpuInfo
= *hwinfo
->cpuInfo
;
131 const gmx::HardwareTopology
&hwTop
= *hwinfo
->hardwareTopology
;
133 s
= gmx::formatString("\n");
134 s
+= gmx::formatString("Running on %d node%s with total",
135 hwinfo
->nphysicalnode
,
136 hwinfo
->nphysicalnode
== 1 ? "" : "s");
137 if (hwinfo
->ncore_tot
> 0)
139 s
+= gmx::formatString(" %d cores,", hwinfo
->ncore_tot
);
141 s
+= gmx::formatString(" %d logical cores", hwinfo
->nhwthread_tot
);
142 if (hwinfo
->gpu_info
.bDetectGPUs
)
144 s
+= gmx::formatString(", %d compatible GPU%s",
145 hwinfo
->ngpu_compatible_tot
,
146 hwinfo
->ngpu_compatible_tot
== 1 ? "" : "s");
150 s
+= gmx::formatString(" (GPU detection deactivated)");
152 s
+= gmx::formatString("\n");
154 if (hwinfo
->nphysicalnode
> 1)
156 /* Print per node hardware feature counts */
157 if (hwinfo
->ncore_max
> 0)
159 s
+= gmx::formatString(" Cores per node: %2d", hwinfo
->ncore_min
);
160 if (hwinfo
->ncore_max
> hwinfo
->ncore_min
)
162 s
+= gmx::formatString(" - %2d", hwinfo
->ncore_max
);
164 s
+= gmx::formatString("\n");
166 s
+= gmx::formatString(" Logical cores per node: %2d", hwinfo
->nhwthread_min
);
167 if (hwinfo
->nhwthread_max
> hwinfo
->nhwthread_min
)
169 s
+= gmx::formatString(" - %2d", hwinfo
->nhwthread_max
);
171 s
+= gmx::formatString("\n");
174 s
+= gmx::formatString(" Compatible GPUs per node: %2d",
175 hwinfo
->ngpu_compatible_min
);
176 if (hwinfo
->ngpu_compatible_max
> hwinfo
->ngpu_compatible_min
)
178 s
+= gmx::formatString(" - %2d", hwinfo
->ngpu_compatible_max
);
180 s
+= gmx::formatString("\n");
181 if (hwinfo
->ngpu_compatible_tot
> 0)
183 if (hwinfo
->bIdenticalGPUs
)
185 s
+= gmx::formatString(" All nodes have identical type(s) of GPUs\n");
189 /* This message will also appear with identical GPU types
190 * when at least one node has no GPU.
192 s
+= gmx::formatString(" Different nodes have different type(s) and/or order of GPUs\n");
202 gmx_gethostname(host
, STRLEN
);
204 MPI_Comm_rank(MPI_COMM_WORLD
, &rank
);
206 // TODO Use a wrapper around MPI_Get_processor_name instead.
207 s
+= gmx::formatString("Hardware detected on host %s (the node of MPI rank %d):\n",
210 s
+= gmx::formatString("Hardware detected:\n");
212 s
+= gmx::formatString(" CPU info:\n");
214 s
+= gmx::formatString(" Vendor: %s\n", cpuInfo
.vendorString().c_str());
216 s
+= gmx::formatString(" Brand: %s\n", cpuInfo
.brandString().c_str());
220 s
+= gmx::formatString(" Family: %d Model: %d Stepping: %d\n",
221 cpuInfo
.family(), cpuInfo
.model(), cpuInfo
.stepping());
223 s
+= gmx::formatString(" Features:");
224 for (auto &f
: cpuInfo
.featureSet())
226 s
+= gmx::formatString(" %s", gmx::CpuInfo::featureString(f
).c_str());
228 s
+= gmx::formatString("\n");
231 if (cpuInfo
.feature(gmx::CpuInfo::Feature::X86_Avx512F
))
233 int avx512fmaunits
= gmx::identifyAvx512FmaUnits();
234 s
+= gmx::formatString(" Number of AVX-512 FMA units:");
235 if (avx512fmaunits
> 0)
237 s
+= gmx::formatString(" %d", avx512fmaunits
);
238 if (avx512fmaunits
== 1)
240 s
+= gmx::formatString(" (AVX2 is faster w/o 2 AVX-512 FMA units)");
245 s
+= gmx::formatString(" Cannot run AVX-512 detection - assuming 2");
247 s
+= gmx::formatString("\n");
250 s
+= gmx::formatString(" Hardware topology: ");
251 switch (hwTop
.supportLevel())
253 case gmx::HardwareTopology::SupportLevel::None
:
254 s
+= gmx::formatString("None\n");
256 case gmx::HardwareTopology::SupportLevel::LogicalProcessorCount
:
257 s
+= gmx::formatString("Only logical processor count\n");
259 case gmx::HardwareTopology::SupportLevel::Basic
:
260 s
+= gmx::formatString("Basic\n");
262 case gmx::HardwareTopology::SupportLevel::Full
:
263 s
+= gmx::formatString("Full\n");
265 case gmx::HardwareTopology::SupportLevel::FullWithDevices
:
266 s
+= gmx::formatString("Full, with devices\n");
270 if (!hwTop
.isThisSystem())
272 s
+= gmx::formatString(" NOTE: Hardware topology cached or synthetic, not detected.\n");
273 if (char *p
= std::getenv("HWLOC_XMLFILE"))
275 s
+= gmx::formatString(" HWLOC_XMLFILE=%s\n", p
);
281 if (hwTop
.supportLevel() >= gmx::HardwareTopology::SupportLevel::Basic
)
283 s
+= gmx::formatString(" Sockets, cores, and logical processors:\n");
285 for (auto &socket
: hwTop
.machine().sockets
)
287 s
+= gmx::formatString(" Socket %2d:", socket
.id
);
288 for (auto &c
: socket
.cores
)
290 s
+= gmx::formatString(" [");
291 for (auto &t
: c
.hwThreads
)
293 s
+= gmx::formatString(" %3d", t
.logicalProcessorId
);
295 s
+= gmx::formatString("]");
297 s
+= gmx::formatString("\n");
300 if (hwTop
.supportLevel() >= gmx::HardwareTopology::SupportLevel::Full
)
302 s
+= gmx::formatString(" Numa nodes:\n");
303 for (auto &n
: hwTop
.machine().numa
.nodes
)
305 s
+= gmx::formatString(" Node %2d (%" PRIu64
" bytes mem):", n
.id
, n
.memory
);
306 for (auto &l
: n
.logicalProcessorId
)
308 s
+= gmx::formatString(" %3d", l
);
310 s
+= gmx::formatString("\n");
312 s
+= gmx::formatString(" Latency:\n ");
313 for (std::size_t j
= 0; j
< hwTop
.machine().numa
.nodes
.size(); j
++)
315 s
+= gmx::formatString(" %5lu", j
);
317 s
+= gmx::formatString("\n");
318 for (std::size_t i
= 0; i
< hwTop
.machine().numa
.nodes
.size(); i
++)
320 s
+= gmx::formatString(" %5lu", i
);
321 for (std::size_t j
= 0; j
< hwTop
.machine().numa
.nodes
.size(); j
++)
323 s
+= gmx::formatString(" %5.2f", hwTop
.machine().numa
.relativeLatency
[i
][j
]);
325 s
+= gmx::formatString("\n");
329 s
+= gmx::formatString(" Caches:\n");
330 for (auto &c
: hwTop
.machine().caches
)
332 s
+= gmx::formatString(" L%d: %" PRIu64
" bytes, linesize %d bytes, assoc. %d, shared %d ways\n",
333 c
.level
, c
.size
, c
.linesize
, c
.associativity
, c
.shared
);
336 if (hwTop
.supportLevel() >= gmx::HardwareTopology::SupportLevel::FullWithDevices
)
338 s
+= gmx::formatString(" PCI devices:\n");
339 for (auto &d
: hwTop
.machine().devices
)
341 s
+= gmx::formatString(" %04x:%02x:%02x.%1x Id: %04x:%04x Class: 0x%04x Numa: %d\n",
342 d
.domain
, d
.bus
, d
.dev
, d
.func
, d
.vendorId
, d
.deviceId
, d
.classId
, d
.numaNodeId
);
347 if (bGPUBinary
&& hwinfo
->gpu_info
.n_dev
> 0)
349 s
+= gmx::formatString(" GPU info:\n");
350 s
+= gmx::formatString(" Number of GPUs detected: %d\n",
351 hwinfo
->gpu_info
.n_dev
);
352 s
+= sprint_gpus(hwinfo
->gpu_info
) + "\n";
357 void gmx_print_detected_hardware(FILE *fplog
, const t_commrec
*cr
,
358 const gmx_multisim_t
*ms
,
359 const gmx::MDLogger
&mdlog
,
360 const gmx_hw_info_t
*hwinfo
)
362 const gmx::CpuInfo
&cpuInfo
= *hwinfo
->cpuInfo
;
364 if (fplog
!= nullptr)
366 std::string detected
;
368 detected
= detected_hardware_string(hwinfo
, TRUE
);
370 fprintf(fplog
, "%s\n", detected
.c_str());
373 // Do not spam stderr with all our internal information unless
374 // there was something that actually went wrong; general information
375 // belongs in the logfile.
377 /* Check the compiled SIMD instruction set against that of the node
378 * with the lowest SIMD level support (skip if SIMD detection did not work)
380 if (cpuInfo
.supportLevel() >= gmx::CpuInfo::SupportLevel::Features
)
382 gmx::simdCheck(static_cast<gmx::SimdType
>(hwinfo
->simd_suggest_min
), fplog
, isMasterSimMasterRank(ms
, cr
));
385 /* For RDTSCP we only check on our local node and skip the MPI reduction */
386 check_use_of_rdtscp_on_this_cpu(mdlog
, cpuInfo
);