| 1 |
#include <cassert> |
| 2 |
#include <fstream> |
| 3 |
#include <algorithm> |
| 4 |
#include "utils/simError.h" |
| 5 |
#include "math/ParallelRandNumGen.hpp" |
| 6 |
|
| 7 |
void RandNumGenTestCase::testUniform(){ |
| 8 |
MTRand randNumGen(823645754); |
| 9 |
|
| 10 |
const int N = 16; |
| 11 |
std::vector<unsigned long int> histogram(N, 0); |
| 12 |
const int num = 1000000; |
| 13 |
for (int i = 0; i <num; ++i) { |
| 14 |
++histogram[randNumGen.randInt(N -1 )]; // rantInt returns an integer in [0, N-1] |
| 15 |
} |
| 16 |
|
| 17 |
std::ofstream uniform("uniform.dat") |
| 18 |
int avg = num / N; |
| 19 |
double tolerance = 0.01*avg; |
| 20 |
for (int i = 0; i < num; ++i) { |
| 21 |
assert((histogram[i] - avg) /avg <= tolerance); |
| 22 |
uniform << i << "\t" << histogram[i] << std::endl; |
| 23 |
} |
| 24 |
} |
| 25 |
|
| 26 |
void RandNumGenTestCase::testGaussian(){ |
| 27 |
MTRand randNumGen(823645754); |
| 28 |
double mean = 100.0; |
| 29 |
double variance = 1.0; |
| 30 |
const int num = 1000000; |
| 31 |
double interval = 0.1; |
| 32 |
const int size = 2000; |
| 33 |
vector<unsigned long int> histogram(size , 0); |
| 34 |
vector<double> normalizedHistogram(size); |
| 35 |
for (int i = 0; i < num; ++i) { |
| 36 |
int index = static_cast<int>(randNumGen.randNorm(mean, variance) / interval); |
| 37 |
++histogram[index]; |
| 38 |
} |
| 39 |
|
| 40 |
std::transform(histogram.begin(), histogram.end(), normalizedHistogram.begin(), std::bind2nd(std::divides<double>(), num)); |
| 41 |
std::ofstream gaussian("gaussian.dat"); |
| 42 |
for (int i = 0; i < num; ++i) { |
| 43 |
gaussian << i << "\t" << normalizedHistogram[i] << std::endl; |
| 44 |
} |
| 45 |
} |
| 46 |
|
| 47 |
void RandNumGenTestCase::testParallelRandNumGen(){ |
| 48 |
const int seed = 324271632; |
| 49 |
const int nloops = 1000000; |
| 50 |
MPI_Status istatus; |
| 51 |
ParallelRandNumGen mpiRandNumGen(seed); |
| 52 |
const int masterNode = 0; |
| 53 |
if (worldRank = masterNode) { |
| 54 |
|
| 55 |
MTRand singleRandNumGen(seed); |
| 56 |
|
| 57 |
int nProcessors; |
| 58 |
MPI_Comm_size(MPI_COMM_WORLD, &nProcessors); |
| 59 |
std::vector<unsigned long int> mpiRandNums(nProcessors); |
| 60 |
std::vector<unsigned long int> singleRandNums(nProcessors); |
| 61 |
|
| 62 |
for (int i = 0; i < nloops; ++i) { |
| 63 |
mpiRandNums[masterNode] = mpiRandNumGen.randInt(); |
| 64 |
|
| 65 |
for (int j = 0; j < nProcessors; ++j) { |
| 66 |
if (j != masterNode) { |
| 67 |
MPI_Recv(&mpiRandNums[j], 1, MPI_UNSIGNED_LONG, j, i, MPI_COMM_WORLD, &istatus); |
| 68 |
} |
| 69 |
|
| 70 |
singleRandNums[j] = mpiRandNumGen.randInt(); |
| 71 |
} |
| 72 |
|
| 73 |
assert(mpiRandNums, singleRandNums); |
| 74 |
} |
| 75 |
|
| 76 |
|
| 77 |
|
| 78 |
} else { |
| 79 |
|
| 80 |
unsigned long int randNum; |
| 81 |
for (int i = 0; i < nloops; ++i) { |
| 82 |
randNum = mpiRandNumGen.randInt(); |
| 83 |
MPI_Send(&randNum, 1, MPI_INT, masterNode, i, MPI_COMM_WORLD); |
| 84 |
} |
| 85 |
|
| 86 |
} |
| 87 |
|
| 88 |
} |
| 89 |
|
| 90 |
|
| 91 |
int main(int argc, char* argv[]) { |
| 92 |
|
| 93 |
MPI_Init(argc, argv); |
| 94 |
|
| 95 |
if (worldRank == 0 ) { |
| 96 |
testUniform(); |
| 97 |
testGaussian(); |
| 98 |
} |
| 99 |
|
| 100 |
testParallelRandNumGen(); |
| 101 |
|
| 102 |
MPI_Finalize(); |
| 103 |
} |
