add speed test

test/Makefile

@@ -1,17 +1,17 @@
 export CC  = gcc
 export CXX = g++
 export MPICXX = mpicxx
-export LDFLAGS= -pthread -lm 
+export LDFLAGS= -pthread -lm  -lrt
 export CFLAGS = -Wall -O3 -msse2  -Wno-unknown-pragmas -fPIC -I../src
 
 # specify tensor path
-BIN = test_allreduce test_recover test_model_recover
+BIN = test_allreduce test_recover test_model_recover speed_test
 # objectives that makes up rabit library
 RABIT_OBJ = allreduce_base.o allreduce_robust.o engine.o
 MPIOBJ = engine_mpi.o
 
-OBJ = $(RABIT_OBJ) test_allreduce.o test_recover.o test_model_recover.o
-MPIBIN = test_allreduce.mpi
+OBJ = $(RABIT_OBJ) test_allreduce.o test_recover.o test_model_recover.o speed_test.o
+MPIBIN = test_allreduce.mpi speed_test.mpi
 .PHONY: clean all
 
 all: $(BIN) $(MPIBIN)
@@ -21,23 +21,26 @@ engine.o: ../src/engine.cc ../src/*.h
 allreduce_robust.o: ../src/allreduce_robust.cc ../src/*.h
 engine_mpi.o: ../src/engine_mpi.cc
 test_allreduce.o: test_allreduce.cpp ../src/*.h
+speed_test.o: speed_test.cpp ../src/*.h
 test_recover.o: test_recover.cpp ../src/*.h 
 test_model_recover.o: test_model_recover.cpp ../src/*.h
 
 # we can link against MPI version to get use MPI
 test_allreduce: test_allreduce.o  $(RABIT_OBJ)
 test_allreduce.mpi: test_allreduce.o $(MPIOBJ)
+speed_test: speed_test.o  $(RABIT_OBJ)
+speed_test.mpi: speed_test.o $(MPIOBJ)
 test_recover: test_recover.o  $(RABIT_OBJ)
 test_model_recover: test_model_recover.o  $(RABIT_OBJ)
 
 $(BIN) : 
-	$(CXX) $(CFLAGS) $(LDFLAGS) -o $@ $(filter %.cpp %.o %.c %.cc, $^)
+	$(CXX) $(CFLAGS) -o $@ $(filter %.cpp %.o %.c %.cc, $^) $(LDFLAGS)
 
 $(OBJ) : 
 	$(CXX) -c $(CFLAGS) -o $@ $(firstword $(filter %.cpp %.c %.cc, $^) )
 
 $(MPIBIN) : 
-	$(MPICXX) $(CFLAGS)  $(LDFLAGS) -o $@ $(filter %.cpp %.o %.c %.cc, $^)
+	$(MPICXX) $(CFLAGS) -o $@ $(filter %.cpp %.o %.c %.cc, $^) $(LDFLAGS)
 
 $(MPIOBJ) : 
 	$(MPICXX) -c $(CFLAGS) -o $@ $(firstword $(filter %.cpp %.c %.cc, $^) )

test/speed_test.cpp

@@ -0,0 +1,96 @@
+#include <rabit.h>
+#include <utils.h>
+#include <timer.h>
+#include <cstdio>
+#include <cstdlib>
+#include <cmath>
+#include <time.h>
+
+using namespace rabit;
+
+double max_tdiff, sum_tdiff, bcast_tdiff, tot_tdiff;
+
+inline void TestMax(size_t n) {
+  int rank = rabit::GetRank();
+  //int nproc = rabit::GetWorldSize();
+
+  std::vector<float> ndata(n);
+  for (size_t i = 0; i < ndata.size(); ++i) {
+    ndata[i] = (i * (rank+1)) % 111;
+  }
+  double tstart = utils::GetTime();
+  rabit::Allreduce<op::Max>(&ndata[0], ndata.size());
+  max_tdiff += utils::GetTime() - tstart;
+}
+
+inline void TestSum(size_t n) {
+  int rank = rabit::GetRank();
+  //int nproc = rabit::GetWorldSize();
+  const int z = 131;
+  std::vector<float> ndata(n);
+  for (size_t i = 0; i < ndata.size(); ++i) {
+    ndata[i] = (i * (rank+1)) % z;
+  }
+  double tstart = utils::GetTime();
+  rabit::Allreduce<op::Sum>(&ndata[0], ndata.size());  
+  sum_tdiff += utils::GetTime() - tstart;
+}
+
+inline void TestBcast(size_t n, int root) {
+  int rank = rabit::GetRank();
+  std::string s; s.resize(n);      
+  for (size_t i = 0; i < n; ++i) {
+    s[i] = char(i % 126 + 1);
+  }
+  std::string res;  
+  if (root == rank) {
+    res = s;
+  }
+  double tstart = utils::GetTime();  
+  rabit::Broadcast(&res, root);
+  bcast_tdiff += utils::GetTime() - tstart;  
+}
+
+inline void PrintStats(const char *name, double tdiff) {
+  int nproc = rabit::GetWorldSize();
+  double tsum = tdiff;
+  rabit::Allreduce<op::Sum>(&tsum, 1);
+  double tavg = tsum / nproc;
+  double tsqr = tdiff - tavg;
+  tsqr *= tsqr;
+  rabit::Allreduce<op::Sum>(&tsqr, 1);
+  double tstd = sqrt(tsqr / nproc);
+  if (rabit::GetRank() == 0) {
+    utils::LogPrintf("%s: mean=%g, std=%g sec\n", name, tavg, tstd);
+  }
+}
+
+int main(int argc, char *argv[]) {
+  if (argc < 3) {
+    printf("Usage: <ndata> <nrepeat>\n");
+    return 0;
+  }
+  srand(0);
+  int n = atoi(argv[1]);
+  int nrep = atoi(argv[2]);
+  utils::Check(nrep >= 1, "need to at least repeat running once");
+  rabit::Init(argc, argv);
+  //int rank = rabit::GetRank();
+  int nproc = rabit::GetWorldSize();
+  std::string name = rabit::GetProcessorName();
+  max_tdiff = sum_tdiff = bcast_tdiff = 0;
+  double tstart = utils::GetTime();
+  for (int i = 0; i < nrep; ++i) {
+    TestMax(n);
+    TestSum(n);
+    TestBcast(n, rand() % nproc);
+  }
+  tot_tdiff = utils::GetTime() - tstart;
+  // use allreduce to get the sum and std of time
+  PrintStats("max_tdiff", max_tdiff);
+  PrintStats("sum_tdiff", sum_tdiff);
+  PrintStats("bcast_tdiff", bcast_tdiff);
+  PrintStats("tot_tdiff", tot_tdiff);
+  rabit::Finalize();
+  return 0;
+}