Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Consistent use of context to specify number of threads. (#8733)

Browse Source 
- Use context in all tests.
- Use context in R.
- Use context in C API DMatrix initialization. (0 threads is used as dft).
This commit is contained in:
Jiaming Yuan
2023-01-30 15:25:31 +08:00
committed by GitHub
parent 21a28f2cc5
commit 3760cede0f
24 changed files with 212 additions and 152 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
tests/cpp/common/test_column_matrix.cc
View File

@@ -1,5 +1,5 @@
/*!
* Copyright 2018-2022 by XGBoost Contributors
/**
* Copyright 2018-2023 by XGBoost Contributors
*/
#include <gtest/gtest.h>
@@ -18,11 +18,10 @@ TEST(DenseColumn, Test) {
for (int32_t max_num_bin : max_num_bins) {
auto dmat = RandomDataGenerator(100, 10, 0.0).GenerateDMatrix();
auto sparse_thresh = 0.2;
GHistIndexMatrix gmat{dmat.get(), max_num_bin, sparse_thresh, false,
common::OmpGetNumThreads(0)};
GHistIndexMatrix gmat{dmat.get(), max_num_bin, sparse_thresh, false, AllThreadsForTest()};
ColumnMatrix column_matrix;
for (auto const& page : dmat->GetBatches<SparsePage>()) {
column_matrix.InitFromSparse(page, gmat, sparse_thresh, common::OmpGetNumThreads(0));
column_matrix.InitFromSparse(page, gmat, sparse_thresh, AllThreadsForTest());
}
ASSERT_GE(column_matrix.GetTypeSize(), last);
ASSERT_LE(column_matrix.GetTypeSize(), kUint32BinsTypeSize);
@@ -65,10 +64,10 @@ TEST(SparseColumn, Test) {
static_cast<int32_t>(std::numeric_limits<uint16_t>::max()) + 2};
for (int32_t max_num_bin : max_num_bins) {
auto dmat = RandomDataGenerator(100, 1, 0.85).GenerateDMatrix();
GHistIndexMatrix gmat{dmat.get(), max_num_bin, 0.5f, false, common::OmpGetNumThreads(0)};
GHistIndexMatrix gmat{dmat.get(), max_num_bin, 0.5f, false, AllThreadsForTest()};
ColumnMatrix column_matrix;
for (auto const& page : dmat->GetBatches<SparsePage>()) {
column_matrix.InitFromSparse(page, gmat, 1.0, common::OmpGetNumThreads(0));
column_matrix.InitFromSparse(page, gmat, 1.0, AllThreadsForTest());
}
common::DispatchBinType(column_matrix.GetTypeSize(), [&](auto dtype) {
using T = decltype(dtype);
@@ -93,10 +92,10 @@ TEST(DenseColumnWithMissing, Test) {
static_cast<int32_t>(std::numeric_limits<uint16_t>::max()) + 2};
for (int32_t max_num_bin : max_num_bins) {
auto dmat = RandomDataGenerator(100, 1, 0.5).GenerateDMatrix();
GHistIndexMatrix gmat(dmat.get(), max_num_bin, 0.2, false, common::OmpGetNumThreads(0));
GHistIndexMatrix gmat(dmat.get(), max_num_bin, 0.2, false, AllThreadsForTest());
ColumnMatrix column_matrix;
for (auto const& page : dmat->GetBatches<SparsePage>()) {
column_matrix.InitFromSparse(page, gmat, 0.2, common::OmpGetNumThreads(0));
column_matrix.InitFromSparse(page, gmat, 0.2, AllThreadsForTest());
}
ASSERT_TRUE(column_matrix.AnyMissing());
DispatchBinType(column_matrix.GetTypeSize(), [&](auto dtype) {

44
tests/cpp/common/test_hist_util.cc
View File

@@ -1,5 +1,5 @@
/*!
* Copyright 2019-2022 by XGBoost Contributors
/**
* Copyright 2019-2023 by XGBoost Contributors
*/
#include <gtest/gtest.h>
#include <vector>
@@ -14,15 +14,13 @@
namespace xgboost {
namespace common {
size_t GetNThreads() { return common::OmpGetNumThreads(0); }
void ParallelGHistBuilderReset() {
constexpr size_t kBins = 10;
constexpr size_t kNodes = 5;
constexpr size_t kNodesExtended = 10;
constexpr size_t kTasksPerNode = 10;
constexpr double kValue = 1.0;
const size_t nthreads = GetNThreads();
const size_t nthreads = AllThreadsForTest();
HistCollection collection;
collection.Init(kBins);
@@ -78,7 +76,7 @@ void ParallelGHistBuilderReduceHist(){
constexpr size_t kNodes = 5;
constexpr size_t kTasksPerNode = 10;
constexpr double kValue = 1.0;
const size_t nthreads = GetNThreads();
const size_t nthreads = AllThreadsForTest();
HistCollection collection;
collection.Init(kBins);
@@ -167,7 +165,7 @@ TEST(HistUtil, DenseCutsCategorical) {
std::vector<float> x_sorted(x);
std::sort(x_sorted.begin(), x_sorted.end());
auto dmat = GetDMatrixFromData(x, n, 1);
HistogramCuts cuts = SketchOnDMatrix(dmat.get(), num_bins, common::OmpGetNumThreads(0));
HistogramCuts cuts = SketchOnDMatrix(dmat.get(), num_bins, AllThreadsForTest());
auto cuts_from_sketch = cuts.Values();
EXPECT_LT(cuts.MinValues()[0], x_sorted.front());
EXPECT_GT(cuts_from_sketch.front(), x_sorted.front());
@@ -180,13 +178,12 @@ TEST(HistUtil, DenseCutsCategorical) {
TEST(HistUtil, DenseCutsAccuracyTest) {
int bin_sizes[] = {2, 16, 256, 512};
int sizes[] = {100};
// omp_set_num_threads(1);
int num_columns = 5;
for (auto num_rows : sizes) {
auto x = GenerateRandom(num_rows, num_columns);
auto dmat = GetDMatrixFromData(x, num_rows, num_columns);
for (auto num_bins : bin_sizes) {
HistogramCuts cuts = SketchOnDMatrix(dmat.get(), num_bins, common::OmpGetNumThreads(0));
HistogramCuts cuts = SketchOnDMatrix(dmat.get(), num_bins, AllThreadsForTest());
ValidateCuts(cuts, dmat.get(), num_bins);
}
}
@@ -203,13 +200,11 @@ TEST(HistUtil, DenseCutsAccuracyTestWeights) {
dmat->Info().weights_.HostVector() = w;
for (auto num_bins : bin_sizes) {
{
HistogramCuts cuts =
SketchOnDMatrix(dmat.get(), num_bins, common::OmpGetNumThreads(0), true);
HistogramCuts cuts = SketchOnDMatrix(dmat.get(), num_bins, AllThreadsForTest(), true);
ValidateCuts(cuts, dmat.get(), num_bins);
}
{
HistogramCuts cuts =
SketchOnDMatrix(dmat.get(), num_bins, common::OmpGetNumThreads(0), false);
HistogramCuts cuts = SketchOnDMatrix(dmat.get(), num_bins, AllThreadsForTest(), false);
ValidateCuts(cuts, dmat.get(), num_bins);
}
}
@@ -231,14 +226,14 @@ void TestQuantileWithHessian(bool use_sorted) {
for (auto num_bins : bin_sizes) {
HistogramCuts cuts_hess =
SketchOnDMatrix(dmat.get(), num_bins, common::OmpGetNumThreads(0), use_sorted, hessian);
SketchOnDMatrix(dmat.get(), num_bins, AllThreadsForTest(), use_sorted, hessian);
for (size_t i = 0; i < w.size(); ++i) {
dmat->Info().weights_.HostVector()[i] = w[i] * hessian[i];
}
ValidateCuts(cuts_hess, dmat.get(), num_bins);
HistogramCuts cuts_wh =
SketchOnDMatrix(dmat.get(), num_bins, common::OmpGetNumThreads(0), use_sorted);
SketchOnDMatrix(dmat.get(), num_bins, AllThreadsForTest(), use_sorted);
ValidateCuts(cuts_wh, dmat.get(), num_bins);
ASSERT_EQ(cuts_hess.Values().size(), cuts_wh.Values().size());
@@ -265,7 +260,7 @@ TEST(HistUtil, DenseCutsExternalMemory) {
dmlc::TemporaryDirectory tmpdir;
auto dmat = GetExternalMemoryDMatrixFromData(x, num_rows, num_columns, tmpdir);
for (auto num_bins : bin_sizes) {
HistogramCuts cuts = SketchOnDMatrix(dmat.get(), num_bins, common::OmpGetNumThreads(0));
HistogramCuts cuts = SketchOnDMatrix(dmat.get(), num_bins, AllThreadsForTest());
ValidateCuts(cuts, dmat.get(), num_bins);
}
}
@@ -285,7 +280,7 @@ TEST(HistUtil, IndexBinBound) {
for (auto max_bin : bin_sizes) {
auto p_fmat = RandomDataGenerator(kRows, kCols, 0).GenerateDMatrix();
GHistIndexMatrix hmat(p_fmat.get(), max_bin, 0.5, false, common::OmpGetNumThreads(0));
GHistIndexMatrix hmat(p_fmat.get(), max_bin, 0.5, false, AllThreadsForTest());
EXPECT_EQ(hmat.index.Size(), kRows*kCols);
EXPECT_EQ(expected_bin_type_sizes[bin_id++], hmat.index.GetBinTypeSize());
}
@@ -308,7 +303,7 @@ TEST(HistUtil, IndexBinData) {
for (auto max_bin : kBinSizes) {
auto p_fmat = RandomDataGenerator(kRows, kCols, 0).GenerateDMatrix();
GHistIndexMatrix hmat(p_fmat.get(), max_bin, 0.5, false, common::OmpGetNumThreads(0));
GHistIndexMatrix hmat(p_fmat.get(), max_bin, 0.5, false, AllThreadsForTest());
uint32_t const* offsets = hmat.index.Offset();
EXPECT_EQ(hmat.index.Size(), kRows*kCols);
switch (max_bin) {
@@ -331,9 +326,8 @@ TEST(HistUtil, IndexBinData) {
void TestSketchFromWeights(bool with_group) {
size_t constexpr kRows = 300, kCols = 20, kBins = 256;
size_t constexpr kGroups = 10;
auto m =
RandomDataGenerator{kRows, kCols, 0}.Device(0).GenerateDMatrix();
common::HistogramCuts cuts = SketchOnDMatrix(m.get(), kBins, common::OmpGetNumThreads(0));
auto m = RandomDataGenerator{kRows, kCols, 0}.Device(0).GenerateDMatrix();
common::HistogramCuts cuts = SketchOnDMatrix(m.get(), kBins, AllThreadsForTest());
MetaInfo info;
Context ctx;
@@ -369,7 +363,7 @@ void TestSketchFromWeights(bool with_group) {
if (with_group) {
m->Info().weights_ = decltype(m->Info().weights_)(); // remove weight
HistogramCuts non_weighted = SketchOnDMatrix(m.get(), kBins, common::OmpGetNumThreads(0));
HistogramCuts non_weighted = SketchOnDMatrix(m.get(), kBins, AllThreadsForTest());
for (size_t i = 0; i < cuts.Values().size(); ++i) {
EXPECT_EQ(cuts.Values()[i], non_weighted.Values()[i]);
}
@@ -388,7 +382,7 @@ void TestSketchFromWeights(bool with_group) {
for (size_t i = 0; i < h_weights.size(); ++i) {
h_weights[i] = static_cast<float>(i + 1) / static_cast<float>(kGroups);
}
HistogramCuts weighted = SketchOnDMatrix(m.get(), kBins, common::OmpGetNumThreads(0));
HistogramCuts weighted = SketchOnDMatrix(m.get(), kBins, AllThreadsForTest());
ValidateCuts(weighted, m.get(), kBins);
}
}
@@ -400,10 +394,10 @@ TEST(HistUtil, SketchFromWeights) {
TEST(HistUtil, SketchCategoricalFeatures) {
TestCategoricalSketch(1000, 256, 32, false, [](DMatrix* p_fmat, int32_t num_bins) {
return SketchOnDMatrix(p_fmat, num_bins, common::OmpGetNumThreads(0));
return SketchOnDMatrix(p_fmat, num_bins, AllThreadsForTest());
});
TestCategoricalSketch(1000, 256, 32, true, [](DMatrix* p_fmat, int32_t num_bins) {
return SketchOnDMatrix(p_fmat, num_bins, common::OmpGetNumThreads(0));
return SketchOnDMatrix(p_fmat, num_bins, AllThreadsForTest());
});
}
} // namespace common

8
tests/cpp/common/test_hist_util.cu
View File

@@ -1,5 +1,5 @@
/*!
* Copyright 2019-2022 by XGBoost Contributors
/**
* Copyright 2019-2023 by XGBoost Contributors
*/
#include <gtest/gtest.h>
#include <thrust/device_vector.h>
@@ -27,7 +27,7 @@ namespace common {
template <typename AdapterT>
HistogramCuts GetHostCuts(AdapterT *adapter, int num_bins, float missing) {
data::SimpleDMatrix dmat(adapter, missing, 1);
HistogramCuts cuts = SketchOnDMatrix(&dmat, num_bins, common::OmpGetNumThreads(0));
HistogramCuts cuts = SketchOnDMatrix(&dmat, num_bins, AllThreadsForTest());
return cuts;
}
@@ -39,7 +39,7 @@ TEST(HistUtil, DeviceSketch) {
auto dmat = GetDMatrixFromData(x, num_rows, num_columns);
auto device_cuts = DeviceSketch(0, dmat.get(), num_bins);
HistogramCuts host_cuts = SketchOnDMatrix(dmat.get(), num_bins, common::OmpGetNumThreads(0));
HistogramCuts host_cuts = SketchOnDMatrix(dmat.get(), num_bins, AllThreadsForTest());
EXPECT_EQ(device_cuts.Values(), host_cuts.Values());
EXPECT_EQ(device_cuts.Ptrs(), host_cuts.Ptrs());

10
tests/cpp/common/test_quantile.cc
View File

@@ -1,5 +1,5 @@
/*!
* Copyright 2020-2022 by XGBoost Contributors
/**
* Copyright 2020-2023 by XGBoost Contributors
*/
#include "test_quantile.h"
@@ -73,7 +73,7 @@ void DoTestDistributedQuantile(size_t rows, size_t cols) {
auto hess = Span<float const>{hessian};
ContainerType<use_column> sketch_distributed(n_bins, m->Info().feature_types.ConstHostSpan(),
column_size, false, OmpGetNumThreads(0));
column_size, false, AllThreadsForTest());
if (use_column) {
for (auto const& page : m->GetBatches<SortedCSCPage>()) {
@@ -94,7 +94,7 @@ void DoTestDistributedQuantile(size_t rows, size_t cols) {
std::for_each(column_size.begin(), column_size.end(), [=](auto& size) { size *= world; });
m->Info().num_row_ = world * rows;
ContainerType<use_column> sketch_on_single_node(n_bins, m->Info().feature_types.ConstHostSpan(),
column_size, false, OmpGetNumThreads(0));
column_size, false, AllThreadsForTest());
m->Info().num_row_ = rows;
for (auto rank = 0; rank < world; ++rank) {
@@ -188,7 +188,7 @@ void TestSameOnAllWorkers() {
.MaxCategory(17)
.Seed(rank + seed)
.GenerateDMatrix();
auto cuts = SketchOnDMatrix(m.get(), n_bins, common::OmpGetNumThreads(0));
auto cuts = SketchOnDMatrix(m.get(), n_bins, AllThreadsForTest());
std::vector<float> cut_values(cuts.Values().size() * world, 0);
std::vector<
typename std::remove_reference_t<decltype(cuts.Ptrs())>::value_type>

10
tests/cpp/common/test_transform_range.cc
View File

@@ -1,5 +1,5 @@
/*!
* Copyright 2018-2022 by XGBoost Contributors
/**
* Copyright 2018-2023 by XGBoost Contributors
*/
#include <gtest/gtest.h>
#include <xgboost/base.h>
@@ -45,7 +45,7 @@ TEST(Transform, DeclareUnifiedTest(Basic)) {
out_vec.Fill(0);
Transform<>::Init(TestTransformRange<bst_float>{},
Range{0, static_cast<Range::DifferenceType>(size)}, common::OmpGetNumThreads(0),
Range{0, static_cast<Range::DifferenceType>(size)}, AllThreadsForTest(),
TRANSFORM_GPU)
.Eval(&out_vec, &in_vec);
std::vector<bst_float> res = out_vec.HostVector();
@@ -61,8 +61,8 @@ TEST(TransformDeathTest, Exception) {
EXPECT_DEATH(
{
Transform<>::Init([](size_t idx, common::Span<float const> _in) { _in[idx + 1]; },
Range(0, static_cast<Range::DifferenceType>(kSize)),
common::OmpGetNumThreads(0), -1)
Range(0, static_cast<Range::DifferenceType>(kSize)), AllThreadsForTest(),
-1)
.Eval(&in_vec);
},
"");