From 617a3d256e6b014d8e17f44b3182bf030221d842 Mon Sep 17 00:00:00 2001
From: Martin Bauer <martin.bauer@fau.de>
Date: Mon, 9 Dec 2019 20:36:55 +0100
Subject: [PATCH] Software Pinning in UniformGridGenerated
---
.../UniformGridGenerated/CMakeLists.txt | 2 +-
.../UniformGridGenerated/Pinning.cpp | 330 ++++++++++++++++++
.../benchmarks/UniformGridGenerated/Pinning.h | 44 +++
.../UniformGridGenerated.cpp | 36 +-
4 files changed, 410 insertions(+), 2 deletions(-)
create mode 100644 apps/benchmarks/UniformGridGenerated/Pinning.cpp
create mode 100644 apps/benchmarks/UniformGridGenerated/Pinning.h
diff --git a/apps/benchmarks/UniformGridGenerated/CMakeLists.txt b/apps/benchmarks/UniformGridGenerated/CMakeLists.txt
index 3204db7a..4e5396f9 100644
--- a/apps/benchmarks/UniformGridGenerated/CMakeLists.txt
+++ b/apps/benchmarks/UniformGridGenerated/CMakeLists.txt
@@ -12,7 +12,7 @@ waLBerla_python_file_generates(UniformGridGenerated.py
foreach(config trt smagorinsky mrt3 mrt entropic_kbc_n4 cumulant )
waLBerla_add_executable ( NAME UniformGridBenchmarkGenerated_${config}
- FILES UniformGridGenerated.cpp UniformGridGenerated.py
+ FILES UniformGridGenerated.cpp Pinning.cpp UniformGridGenerated.py
DEPENDS blockforest boundary core domain_decomposition field geometry timeloop vtk gui
CODEGEN_CFG ${config})
endforeach()
diff --git a/apps/benchmarks/UniformGridGenerated/Pinning.cpp b/apps/benchmarks/UniformGridGenerated/Pinning.cpp
new file mode 100644
index 00000000..1de023c5
--- /dev/null
+++ b/apps/benchmarks/UniformGridGenerated/Pinning.cpp
@@ -0,0 +1,330 @@
+// --------------------------------------------------------------------------
+//
+// Copyright
+// Markus Wittmann, 2016-2017
+// RRZE, University of Erlangen-Nuremberg, Germany
+// markus.wittmann -at- fau.de or hpc -at- rrze.fau.de
+//
+// Viktor Haag, 2016
+// LSS, University of Erlangen-Nuremberg, Germany
+//
+// This file is part of the Lattice Boltzmann Benchmark Kernels (LbmBenchKernels).
+//
+// LbmBenchKernels is free software: you can redistribute it and/or modify
+// it under the terms of the GNU General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// LbmBenchKernels is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU General Public License for more details.
+//
+// You should have received a copy of the GNU General Public License
+// along with LbmBenchKernels. If not, see <http://www.gnu.org/licenses/>.
+//
+// --------------------------------------------------------------------------
+#ifndef _GNU_SOURCE
+ #define _GNU_SOURCE
+#endif
+#include <sched.h>
+#include <errno.h>
+
+#include "Pinning.h"
+
+#include "core/logging/Logging.h"
+
+// -----------------------------------------------------------------------
+//
+// Binds the calling thread to specified core.
+//
+// Return value: 0 = success, else error.
+//
+// -----------------------------------------------------------------------
+
+int PinCurrentThreadToCore(int coreNumber)
+{
+ int error = 0;
+
+ cpu_set_t cpu_set;
+ CPU_ZERO(&cpu_set);
+ CPU_SET(coreNumber, &cpu_set);
+
+ error = sched_setaffinity((pid_t)0, sizeof(cpu_set_t), &cpu_set);
+
+ if (error != 0) {
+ WALBERLA_LOG_WARNING("pinning thread to core " << coreNumber << " failed (" << error << "):"<< strerror(error));
+ }
+
+ return error;
+}
+
+
+// -----------------------------------------------------------------------
+//
+// Binds the calling thread to specified core by a cpu list specified
+// in the given environment variable.
+//
+// Return value: 0 = success, else error.
+//
+// -----------------------------------------------------------------------
+
+int PinCurrentThreadByEnvVar(const char * envVarName, int threadNumber)
+{
+ const char * envVarValue;
+ int core;
+
+ envVarValue = getenv(envVarName);
+
+ if (envVarValue == NULL) {
+ WALBERLA_LOG_WARNING("skip pinning: env var " << envVarName << " not set");
+ return 0;
+ }
+
+ core = PinParseCpuList(envVarValue, threadNumber);
+
+ if (core < 0) {
+ return core;
+ }
+
+ return PinCurrentThreadToCore(core);
+}
+
+
+// -----------------------------------------------------------------------
+//
+// Binds the calling thread to a core specified in the CPU list.
+//
+// Return value: 0 = success, else error.
+//
+// -----------------------------------------------------------------------
+
+int PinCurrentThreadByCpuList(const char * cpuList, int threadNumber)
+{
+ int core;
+
+ if (cpuList == NULL) {
+ WALBERLA_ABORT("cpu list is NULL");
+ }
+
+ core = PinParseCpuList(cpuList, threadNumber);
+
+ if (core < 0) {
+ return core;
+ }
+
+ return PinCurrentThreadToCore(core);
+}
+
+
+// -----------------------------------------------------------------------
+//
+// Parses the provided cpu list and returns the core number for the
+// specified thread.
+//
+// The cpu list has for example a format of: 0,1,2
+//
+// -----------------------------------------------------------------------
+
+int PinParseCpuList(const char * cpuList, int threadNumber)
+{
+ int cpu = -1;
+
+ if (cpuList == NULL) {
+ return -1;
+ }
+
+ const char * c = cpuList;
+
+ // Ensure only valid characters are in the cpu list.
+ // Cpu list is in the format of "0,1,2,3,4,5".
+ while (((*c >= '0' && *c <= '9') || *c == ',')) {
+ ++c;
+ }
+
+ if (*c != 0x00) {
+ // Invalid character detected.
+ return -2;
+ }
+
+ c = cpuList;
+
+ // Now find the core for the specified thread.
+
+ int t = 0;
+
+ while (t < threadNumber && *c != 0x00) {
+ if (*c == ',') {
+ ++t;
+ }
+ ++c;
+ }
+
+ if (t != threadNumber || *c < '0' || *c > '9') {
+ // Cpu for this threadNumber not found.
+ return -4;
+ }
+
+ cpu = atoi(c);
+
+ return cpu;
+}
+
+
+
+// -----------------------------------------------------------------------
+//
+// Returns the first core from the calling thread's affinity set.
+//
+// On error a value < 0 is returned.
+//
+// -----------------------------------------------------------------------
+
+int PinCurrentCore()
+{
+ int core = -1;
+ int err;
+
+ cpu_set_t cpu_set;
+ CPU_ZERO(&cpu_set);
+
+ err = sched_getaffinity((pid_t)0, sizeof(cpu_set_t), &cpu_set);
+
+ // constant CPU_SETSIZE is one larger than the maximum CPU
+ // number that can be stored in a CPU set
+ for (int i = 0; i < CPU_SETSIZE; ++i) {
+ if (CPU_ISSET(i, &cpu_set)) {
+ core = i;
+ break;
+ }
+ }
+
+ if (err != 0) {
+ WALBERLA_LOG_WARNING("getting thread affinity failed (" << err << "): " << strerror(err));
+ return -1;
+ }
+
+ return core;
+}
+
+
+
+// -----------------------------------------------------------------------
+//
+// Returns the all cores from the calling thread's affinity set.
+//
+// On error a value < 0 is returned.
+//
+// -----------------------------------------------------------------------
+
+typedef cpu_set_t CpuSet;
+
+
+static CpuSet PinCurrentCores()
+{
+ CpuSet cpuSet;
+ int err;
+
+ cpu_set_t cpu_set;
+ CPU_ZERO(&cpu_set);
+
+ err = sched_getaffinity((pid_t)0, sizeof(cpu_set_t), &cpu_set);
+
+ cpuSet = cpu_set;
+
+ if (err != 0) {
+ WALBERLA_LOG_WARNING("getting thread affinity failed (" << err << "): " << strerror(err));
+ return cpuSet;
+ }
+
+ return cpuSet;
+}
+
+static char * CpuSetToString(cpu_set_t * cpu_set)
+{
+ int previousSetCore = -2;
+ int rangeBeginCore = -2;
+
+ char * buffer1 = (char *)malloc(1024);
+ WALBERLA_ASSERT(buffer1 != NULL);
+ char * buffer2 = (char *)malloc(1024);
+ WALBERLA_ASSERT(buffer2 != NULL);
+
+ buffer1[0] = 0x00;
+ buffer2[0] = 0x00;
+
+ char * buffer = buffer1;
+ char * bufferOld = buffer2;
+
+ const char * empty = "";
+ const char * realComma = ",";
+ const char * comma = empty;
+
+ // TODO: use snprintf
+ // TODO: increase allocated buffer if necessary
+
+ for (int i = 0; i < CPU_SETSIZE; ++i) {
+ if (!CPU_ISSET(i, cpu_set)) {
+ continue;
+ }
+
+ if (i == previousSetCore + 1) {
+ previousSetCore = i;
+ continue;
+ }
+
+ // Now we reached the end of a range.
+ // The range can also consist of only one core.
+ // Be aware, that this core is not part of the range.
+
+ // TODO: this code is repeated below -> use it only once
+ if (rangeBeginCore >= 0 && previousSetCore >= 0) {
+ char * tmp;
+
+ tmp = buffer;
+ buffer = bufferOld;
+ bufferOld = tmp;
+
+ if (rangeBeginCore < previousSetCore) {
+ sprintf(buffer, "%s%s%d-%d", bufferOld, comma, rangeBeginCore, previousSetCore);
+ }
+ else {
+ sprintf(buffer, "%s%s%d", bufferOld, comma, previousSetCore);
+ }
+
+ comma = realComma;
+ }
+
+ // With this core a new range begins.
+ rangeBeginCore = i;
+ previousSetCore = i;
+ }
+
+ if (rangeBeginCore >= 0 && previousSetCore >= 0) {
+ char * tmp;
+
+ tmp = buffer;
+ buffer = bufferOld;
+ bufferOld = tmp;
+
+ if (rangeBeginCore < previousSetCore) {
+ sprintf(buffer, "%s%s%d-%d", bufferOld, comma, rangeBeginCore, previousSetCore);
+ }
+ else {
+ sprintf(buffer, "%s%s%d", bufferOld, comma, previousSetCore);
+ }
+ }
+
+ free(bufferOld); bufferOld = NULL;
+
+ return buffer;
+}
+
+char * PinCpuListAsString()
+{
+ CpuSet cpuSet = PinCurrentCores();
+
+ return CpuSetToString(&cpuSet);
+}
+
+
diff --git a/apps/benchmarks/UniformGridGenerated/Pinning.h b/apps/benchmarks/UniformGridGenerated/Pinning.h
new file mode 100644
index 00000000..877048de
--- /dev/null
+++ b/apps/benchmarks/UniformGridGenerated/Pinning.h
@@ -0,0 +1,44 @@
+// --------------------------------------------------------------------------
+//
+// Copyright
+// Markus Wittmann, 2016-2017
+// RRZE, University of Erlangen-Nuremberg, Germany
+// markus.wittmann -at- fau.de or hpc -at- rrze.fau.de
+//
+// Viktor Haag, 2016
+// LSS, University of Erlangen-Nuremberg, Germany
+//
+// This file is part of the Lattice Boltzmann Benchmark Kernels (LbmBenchKernels).
+//
+// LbmBenchKernels is free software: you can redistribute it and/or modify
+// it under the terms of the GNU General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// LbmBenchKernels is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU General Public License for more details.
+//
+// You should have received a copy of the GNU General Public License
+// along with LbmBenchKernels. If not, see <http://www.gnu.org/licenses/>.
+//
+// --------------------------------------------------------------------------
+#ifndef __PINNING_H__
+#define __PINNING_H__
+
+
+int PinCurrentThreadToCore(int coreNumber);
+
+int PinParseCpuList(const char * cpuList, int threadNumber);
+
+int PinCurrentThreadByEnvVar(const char * envVarName, int threadNumber);
+
+int PinCurrentThreadByCpuList(const char * cpuList, int threadNumber);
+
+int PinCurrentCore();
+
+char * PinCpuListAsString();
+
+
+#endif // __PINNING_H__
diff --git a/apps/benchmarks/UniformGridGenerated/UniformGridGenerated.cpp b/apps/benchmarks/UniformGridGenerated/UniformGridGenerated.cpp
index 28707b21..e1b013e0 100644
--- a/apps/benchmarks/UniformGridGenerated/UniformGridGenerated.cpp
+++ b/apps/benchmarks/UniformGridGenerated/UniformGridGenerated.cpp
@@ -33,7 +33,7 @@
#include "GenMpiDtypeInfo.h"
#include "GenMpiDtypeInfoAAPull.h"
#include "GenMpiDtypeInfoAAPush.h"
-
+#include "Pinning.h"
#include <iomanip>
@@ -43,6 +43,36 @@ using PdfField_T = GhostLayerField< real_t, Stencil_T::Q >;
using VelocityField_T = GhostLayerField< real_t, 3 >;
+void pinOpenMP(const char * pinString)
+{
+#ifdef WALBERLA_BUILD_WITH_OPENMP
+ if (pinString != NULL) {
+ #pragma omp parallel
+ {
+ int threadId = omp_get_thread_num();
+ int err;
+
+ err = PinCurrentThreadByCpuList(pinString, threadId);
+
+ if (err) {
+ WALBERLA_ABORT("Pinning of " << threadId << "failed");
+ }
+
+ const char * cpuList = PinCpuListAsString();
+ WALBERLA_ASSERT(cpuList != NULL);
+
+ // Not so nice hack to print the thread ids ordered.
+ #pragma omp for ordered
+ for (int i = 0; i < omp_get_num_threads(); ++i) {
+ #pragma omp ordered
+ WALBERLA_LOG_INFO("Thread " << threadId << " pinned to core(s) " << cpuList);
+ }
+ free((void *)cpuList);
+ }
+ }
+#endif
+}
+
int main( int argc, char **argv )
{
mpi::Environment env( argc, argv );
@@ -64,6 +94,10 @@ int main( int argc, char **argv )
const real_t shearVelocityMagnitude = parameters.getParameter<real_t>("shearVelocityMagnitude", 0.02);
const bool directComm = parameters.getParameter<bool>("directComm", false);
+ const std::string pinning = parameters.getParameter<std::string>("pinning", "");
+ if( !pinning.empty() )
+ pinOpenMP(pinning.c_str());
+
auto pdfFieldAdder = [](IBlock* const block, StructuredBlockStorage * const storage) {
return new PdfField_T(storage->getNumberOfXCells(*block),
storage->getNumberOfYCells(*block),
--
GitLab