Expose RabitAllGatherRing and RabitGetRingPrevRank (#113)

* add unittests

* Expose RabitAllGatherRing and RabitGetRingPrevRank

* Enabled TCP_NODELAY to decrease latency
This commit is contained in:
nateagr
2019-11-12 12:55:32 +01:00
committed by Jiaming Yuan
parent 90e2239372
commit 1907b25cd0
15 changed files with 509 additions and 25 deletions

View File

@@ -0,0 +1,66 @@
#define RABIT_CXXTESTDEFS_H
#include <gtest/gtest.h>
#include <string>
#include <iostream>
#include "../../src/allreduce_base.h"
TEST(allreduce_base, init_task)
{
rabit::engine::AllreduceBase base;
std::string rabit_task_id = "rabit_task_id=1";
char cmd[rabit_task_id.size()+1];
std::copy(rabit_task_id.begin(), rabit_task_id.end(), cmd);
cmd[rabit_task_id.size()] = '\0';
char* argv[] = {cmd};
base.Init(1, argv);
EXPECT_EQ(base.task_id, "1");
}
TEST(allreduce_base, init_with_cache_on)
{
rabit::engine::AllreduceBase base;
std::string rabit_task_id = "rabit_task_id=1";
char cmd[rabit_task_id.size()+1];
std::copy(rabit_task_id.begin(), rabit_task_id.end(), cmd);
cmd[rabit_task_id.size()] = '\0';
std::string rabit_bootstrap_cache = "rabit_bootstrap_cache=1";
char cmd2[rabit_bootstrap_cache.size()+1];
std::copy(rabit_bootstrap_cache.begin(), rabit_bootstrap_cache.end(), cmd2);
cmd2[rabit_bootstrap_cache.size()] = '\0';
std::string rabit_debug = "rabit_debug=1";
char cmd3[rabit_debug.size()+1];
std::copy(rabit_debug.begin(), rabit_debug.end(), cmd3);
cmd3[rabit_debug.size()] = '\0';
char* argv[] = {cmd, cmd2, cmd3};
base.Init(3, argv);
EXPECT_EQ(base.task_id, "1");
EXPECT_EQ(base.rabit_bootstrap_cache, 1);
EXPECT_EQ(base.rabit_debug, 1);
}
TEST(allreduce_base, init_with_ring_reduce)
{
rabit::engine::AllreduceBase base;
std::string rabit_task_id = "rabit_task_id=1";
char cmd[rabit_task_id.size()+1];
std::copy(rabit_task_id.begin(), rabit_task_id.end(), cmd);
cmd[rabit_task_id.size()] = '\0';
std::string rabit_reduce_ring_mincount = "rabit_reduce_ring_mincount=1";
char cmd2[rabit_reduce_ring_mincount.size()+1];
std::copy(rabit_reduce_ring_mincount.begin(), rabit_reduce_ring_mincount.end(), cmd2);
cmd2[rabit_reduce_ring_mincount.size()] = '\0';
char* argv[] = {cmd, cmd2};
base.Init(2, argv);
EXPECT_EQ(base.task_id, "1");
EXPECT_EQ(base.reduce_ring_mincount, 1);
}

View File

@@ -0,0 +1,51 @@
#define RABIT_CXXTESTDEFS_H
#include <gtest/gtest.h>
#include <string>
#include <iostream>
#include "../../src/allreduce_mock.h"
TEST(allreduce_mock, mock_allreduce)
{
rabit::engine::AllreduceMock m;
std::string mock_str = "mock=0,0,0,0";
char cmd[mock_str.size()+1];
std::copy(mock_str.begin(), mock_str.end(), cmd);
cmd[mock_str.size()] = '\0';
char* argv[] = {cmd};
m.Init(1, argv);
m.rank = 0;
EXPECT_EXIT(m.Allreduce(nullptr,0,0,nullptr,nullptr,nullptr), ::testing::ExitedWithCode(255), "");
}
TEST(allreduce_mock, mock_broadcast)
{
rabit::engine::AllreduceMock m;
std::string mock_str = "mock=0,1,2,0";
char cmd[mock_str.size()+1];
std::copy(mock_str.begin(), mock_str.end(), cmd);
cmd[mock_str.size()] = '\0';
char* argv[] = {cmd};
m.Init(1, argv);
m.rank = 0;
m.version_number=1;
m.seq_counter=2;
EXPECT_EXIT(m.Broadcast(nullptr,0,0), ::testing::ExitedWithCode(255), "");
}
TEST(allreduce_mock, mock_gather)
{
rabit::engine::AllreduceMock m;
std::string mock_str = "mock=3,13,22,0";
char cmd[mock_str.size()+1];
std::copy(mock_str.begin(), mock_str.end(), cmd);
cmd[mock_str.size()] = '\0';
char* argv[] = {cmd};
m.Init(1, argv);
m.rank = 3;
m.version_number=13;
m.seq_counter=22;
EXPECT_EXIT(m.Allgather(nullptr,0,0,0,0), ::testing::ExitedWithCode(255), "");
}

View File

@@ -26,7 +26,7 @@ class Model : public rabit::Serializable {
}
};
inline void TestMax(Model *model, int ntrial, int iter) {
inline void TestMax(Model *model, int iter) {
int rank = rabit::GetRank();
int nproc = rabit::GetWorldSize();
const int z = iter + 111;
@@ -47,7 +47,7 @@ inline void TestMax(Model *model, int ntrial, int iter) {
model->data = ndata;
}
inline void TestSum(Model *model, int ntrial, int iter) {
inline void TestSum(Model *model, int iter) {
int rank = rabit::GetRank();
int nproc = rabit::GetWorldSize();
const int z = 131 + iter;
@@ -69,7 +69,30 @@ inline void TestSum(Model *model, int ntrial, int iter) {
model->data = ndata;
}
inline void TestBcast(size_t n, int root, int ntrial, int iter) {
inline void TestAllgather(Model *model, int iter) {
int rank = rabit::GetRank();
int nproc = rabit::GetWorldSize();
const int z = 131 + iter;
std::vector<float> ndata(model->data.size() * nproc);
size_t beginSlice = rank * model->data.size();
for (size_t i = 0; i < model->data.size(); ++i) {
ndata[beginSlice + i] = (i * (rank+1)) % z + model->data[i];
}
Allgather(&ndata[0], ndata.size(), beginSlice,
model->data.size(), model->data.size());
for (size_t i = 0; i < ndata.size(); ++i) {
int curRank = i / model->data.size();
int remainder = i % model->data.size();
float data = (remainder * (curRank+1)) % z + model->data[remainder];
utils::Check(fabsf(data - ndata[i]) < 1e-5 ,
"[%d] TestAllgather check failure, local=%g, allgatherring=%g", rank, data, ndata[i]);
}
model->data = ndata;
}
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) {
@@ -113,19 +136,22 @@ int main(int argc, char *argv[]) {
printf("[%d] reload-trail=%d, init iter=%d\n", rank, ntrial, iter);
for (int r = iter; r < 3; ++r) {
TestMax(&model, ntrial, r);
TestMax(&model, r);
printf("[%d] !!!TestMax pass, iter=%d\n", rank, r);
int step = std::max(nproc / 3, 1);
for (int i = 0; i < nproc; i += step) {
TestBcast(n, i, ntrial, r);
TestBcast(n, i);
}
printf("[%d] !!!TestBcast pass, iter=%d\n", rank, r);
TestSum(&model, ntrial, r);
TestSum(&model, r);
printf("[%d] !!!TestSum pass, iter=%d\n", rank, r);
TestAllgather(&model, r);
printf("[%d] !!!TestAllgather pass, iter=%d\n", rank, r);
rabit::CheckPoint(&model);
printf("[%d] !!!Checkpoint pass, iter=%d\n", rank, r);
}
rabit::Finalize();
return 0;
}