1. fix core bind bugs

benchmark/src/Benchmark.cpp

@@ -22,6 +22,7 @@ void runSingleThreadTest(std::string configName) {
   uint64_t usingMeter = cfg->tryU64("usingMeter", 0, true);
   std::string meterTag = cfg->tryString("meterTag", "intelMsr", true);
   uint64_t useCPP = cfg->tryU64("useCPP", 0, true);
+  uint64_t  forceMP=cfg->tryU64("forceMP", 0, true);
   if (usingMeter) {
     eMeter = meterTable.findMeter(meterTag);
     if (eMeter != nullptr) {
@@ -67,7 +68,7 @@ auto B = torch::rand({(long) aCol, (long) bCol});*/
   ThreadPerf pef(-1);
   pef.setPerfList();
   AMMBench::BlockPartitionRunner br;
-  if (threads > 1) {
+  if (threads > 1||forceMP) {
     INTELLI_WARNING("use multithread");
     br.setConfig(cfg);
     br.createABC(A, B);
@@ -113,9 +114,10 @@ auto B = torch::rand({(long) aCol, (long) bCol});*/
     resultCsv->edit("energyAll", (double) energyConsumption);
     resultCsv->edit("energyOnlyMe", (double) pureEnergy);
   }
-  if (threads > 1) {
+  if (threads > 1||forceMP) {
     INTELLI_WARNING("consider multithread elapsed time");
     resultCsv->edit("perfElapsedTime", (uint64_t) br.getElapsedTime());
+    br.appendThreadInfo(resultCsv);
   }
   // error
   INTELLI_WARNING("evaluating the error, may takes some time");

commit.sh

@@ -1,4 +1,4 @@
-BRANCH=count-sketch-cpp
+BRANCH=core-bind-bug
 git init
 git checkout -b $BRANCH
 git add .

commit_info

@@ -1 +1 @@
-1. FIX CONFLICTS WITH MAIN
+1. fix core bind bugs

include/Parallelization/BlockPartitionRunner.h

@@ -74,6 +74,10 @@ class BlockPartitionWorker : public INTELLI::AbstractC20Thread {
    * @param mycore the core to be binded
    */
   void setWorkParameters(uint64_t aStart, uint64_t aEnd, int mycore);
+  void setCoreBInd(int cno)
+  {
+    coreBind=cno;
+  }
   ~BlockPartitionWorker() {
 
   }
@@ -118,6 +122,10 @@ class BlockPartitionRunner {
   */
   TensorPtr matC = nullptr;  // Output matrix C
   std::vector<BlockPartitionWorkerPtr> workers;
+  /**
+   * @brief special bind of first core, if need
+   */
+  uint64_t firstCoreBind=0;
  public:
   BlockPartitionRunner() {}
   ~BlockPartitionRunner() {}
@@ -154,6 +162,11 @@ class BlockPartitionRunner {
    * @note Exclude the overhead of cleaning thread states such as loaded module
    */
   uint64_t getElapsedTime();
+  /**
+   * @brief append the running information of each thread to the result csv
+   * @param ru The result csv to be appended
+   */
+   void appendThreadInfo(INTELLI::ConfigMapPtr ru);
 };
 
 } // AMMBench

src/Parallelization/BlockPartitionRunner.cpp

@@ -71,10 +71,12 @@ void AMMBench::BlockPartitionRunner::setConfig(INTELLI::ConfigMapPtr _cfg) {
   cfg = _cfg;
   threads = cfg->tryU64("threads", 2, true);
   workers = std::vector<BlockPartitionWorkerPtr>(threads);
+  firstCoreBind=cfg->tryU64("firstCoreBind",0, false);
   for (uint64_t i = 0; i < threads; i++) {
     workers[i] = newBlockPartitionWorker();
     workers[i]->setConfig(cfg);
   }
+
   INTELLI_INFO("set up " + to_string(threads) + "workers.");
 }
 void AMMBench::BlockPartitionRunner::createABC(torch::Tensor A, torch::Tensor B) {
@@ -89,6 +91,11 @@ void AMMBench::BlockPartitionRunner::createABC(torch::Tensor A, torch::Tensor B)
     workers[i]->setABC(matA, matB, matC);
     workers[i]->setWorkParameters(start_row, end_row, i);
   }
+  if(firstCoreBind!=0)
+  {
+    workers[0]->setCoreBInd((int)firstCoreBind);
+    INTELLI_INFO("first thread is bound to core" + to_string(firstCoreBind) );
+  }
 }
 torch::Tensor AMMBench::BlockPartitionRunner::parallelForward() {
   for (uint64_t i = 0; i < threads; i++) {
@@ -114,4 +121,10 @@ uint64_t AMMBench::BlockPartitionRunner::getElapsedTime() {
     ti += workers[i]->getElapsedTime();
   }
   return ti / threads;
+}
+void AMMBench::BlockPartitionRunner::appendThreadInfo(INTELLI::ConfigMapPtr ru) {
+  for (uint64_t i = 0; i < threads; i++) {
+    std::string keyElapesedTime="thread"+ to_string(i)+"RunTime";
+    ru->edit(keyElapesedTime,(uint64_t) workers[i]->getElapsedTime());
+  }
 }

src/Utils/UtilityFunctions.cpp

@@ -4,6 +4,11 @@
 #include <iostream>
 #include <numeric>
 
+#include <time.h>
+#include <sched.h>
+#include <pthread.h>
+#include<cstdlib>
+#include<time.h>
 using namespace std;
 
 /*
@@ -27,6 +32,13 @@ int INTELLI::UtilityFunctions::bind2Core(int id) {
   cpu_set_t mask;
   CPU_ZERO(&mask);
   CPU_SET(cpuId, &mask);
+  /**
+   * @brief fixed some core bind bugs
+   */
+  if (sched_setaffinity(0, sizeof(cpu_set_t), &mask) < 0) {
+    AT_ERROR("Error: setaffinity()\n");
+    exit(0);
+  }
   return cpuId;
 }