Remove omp_get_max_threads (#7608)

This is the one last PR for removing omp global variable.

* Add context object to the `DMatrix`.  This bridges `DMatrix` with https://github.com/dmlc/xgboost/issues/7308 .
* Require context to be available at the construction time of booster.
* Add `n_threads` support for R csc DMatrix constructor.
* Remove `omp_get_max_threads` in R glue code.
* Remove threading utilities that rely on omp global variable.

tests/cpp/predictor/test_cpu_predictor.cc

@@ -25,7 +25,9 @@ TEST(CpuPredictor, Basic) {
   param.base_score = 0.0;
   param.num_output_group = 1;
 
-  gbm::GBTreeModel model = CreateTestModel(&param);
+  GenericParameter ctx;
+  ctx.UpdateAllowUnknown(Args{});
+  gbm::GBTreeModel model = CreateTestModel(&param, &ctx);
 
   auto dmat = RandomDataGenerator(kRows, kCols, 0).GenerateDMatrix();
 
@@ -106,7 +108,9 @@ TEST(CpuPredictor, ExternalMemory) {
   param.num_feature = dmat->Info().num_col_;
   param.num_output_group = 1;
 
-  gbm::GBTreeModel model = CreateTestModel(&param);
+  GenericParameter ctx;
+  ctx.UpdateAllowUnknown(Args{});
+  gbm::GBTreeModel model = CreateTestModel(&param, &ctx);
 
   // Test predict batch
   PredictionCacheEntry out_predictions;

tests/cpp/predictor/test_gpu_predictor.cu

@@ -38,7 +38,9 @@ TEST(GPUPredictor, Basic) {
     param.num_output_group = 1;
     param.base_score = 0.5;
 
-    gbm::GBTreeModel model = CreateTestModel(&param);
+    GenericParameter ctx;
+    ctx.UpdateAllowUnknown(Args{});
+    gbm::GBTreeModel model = CreateTestModel(&param, &ctx);
 
     // Test predict batch
     PredictionCacheEntry gpu_out_predictions;
@@ -100,7 +102,9 @@ TEST(GPUPredictor, ExternalMemoryTest) {
   param.num_output_group = n_classes;
   param.base_score = 0.5;
 
-  gbm::GBTreeModel model = CreateTestModel(&param, n_classes);
+  GenericParameter ctx;
+  ctx.UpdateAllowUnknown(Args{});
+  gbm::GBTreeModel model = CreateTestModel(&param, &ctx, n_classes);
   std::vector<std::unique_ptr<DMatrix>> dmats;
 
   dmats.push_back(CreateSparsePageDMatrix(400));
@@ -167,11 +171,17 @@ TEST(GpuPredictor, LesserFeatures) {
 // Very basic test of empty model
 TEST(GPUPredictor, ShapStump) {
   cudaSetDevice(0);
+
   LearnerModelParam param;
   param.num_feature = 1;
   param.num_output_group = 1;
   param.base_score = 0.5;
-  gbm::GBTreeModel model(&param);
+
+  GenericParameter ctx;
+  ctx.UpdateAllowUnknown(Args{});
+
+  gbm::GBTreeModel model(&param, &ctx);
+
   std::vector<std::unique_ptr<RegTree>> trees;
   trees.push_back(std::unique_ptr<RegTree>(new RegTree));
   model.CommitModel(std::move(trees), 0);
@@ -197,7 +207,12 @@ TEST(GPUPredictor, Shap) {
   param.num_feature = 1;
   param.num_output_group = 1;
   param.base_score = 0.5;
-  gbm::GBTreeModel model(&param);
+
+  GenericParameter ctx;
+  ctx.UpdateAllowUnknown(Args{});
+
+  gbm::GBTreeModel model(&param, &ctx);
+
   std::vector<std::unique_ptr<RegTree>> trees;
   trees.push_back(std::unique_ptr<RegTree>(new RegTree));
   trees[0]->ExpandNode(0, 0, 0.5, true, 1.0, -1.0, 1.0, 0.0, 5.0, 2.0, 3.0);
@@ -249,7 +264,9 @@ TEST(GPUPredictor, PredictLeafBasic) {
   param.base_score = 0.0;
   param.num_output_group = 1;
 
-  gbm::GBTreeModel model = CreateTestModel(&param);
+  GenericParameter ctx;
+  ctx.UpdateAllowUnknown(Args{});
+  gbm::GBTreeModel model = CreateTestModel(&param, &ctx);
 
   HostDeviceVector<float> leaf_out_predictions;
   gpu_predictor->PredictLeaf(dmat.get(), &leaf_out_predictions, model);

tests/cpp/predictor/test_predictor.cc

@@ -214,10 +214,11 @@ void TestCategoricalPrediction(std::string name) {
   float left_weight = 1.3f;
   float right_weight = 1.7f;
 
-  gbm::GBTreeModel model(&param);
+  GenericParameter ctx;
+  ctx.UpdateAllowUnknown(Args{});
+  gbm::GBTreeModel model(&param, &ctx);
   GBTreeModelForTest(&model, split_ind, split_cat, left_weight, right_weight);
 
-  GenericParameter ctx;
   ctx.UpdateAllowUnknown(Args{{"gpu_id", "0"}});
   std::unique_ptr<Predictor> predictor{Predictor::Create(name.c_str(), &ctx)};
 
@@ -257,13 +258,14 @@ void TestCategoricalPredictLeaf(StringView name) {
   float left_weight = 1.3f;
   float right_weight = 1.7f;
 
-  gbm::GBTreeModel model(&param);
+  GenericParameter ctx;
+  ctx.UpdateAllowUnknown(Args{});
+
+  gbm::GBTreeModel model(&param, &ctx);
   GBTreeModelForTest(&model, split_ind, split_cat, left_weight, right_weight);
 
-  GenericParameter runtime;
-  runtime.gpu_id = 0;
-  std::unique_ptr<Predictor> predictor{
-      Predictor::Create(name.c_str(), &runtime)};
+  ctx.gpu_id = 0;
+  std::unique_ptr<Predictor> predictor{Predictor::Create(name.c_str(), &ctx)};
 
   std::vector<float> row(kCols);
   row[split_ind] = split_cat;

tests/cpp/predictor/test_predictor.h

@@ -23,7 +23,9 @@ void TestPredictionFromGradientIndex(std::string name, size_t rows, size_t cols,
       std::unique_ptr<Predictor>(Predictor::Create(name, &lparam));
   predictor->Configure({});
 
-  gbm::GBTreeModel model = CreateTestModel(&param, kClasses);
+  GenericParameter ctx;
+  ctx.UpdateAllowUnknown(Args{});
+  gbm::GBTreeModel model = CreateTestModel(&param, &ctx, kClasses);
 
   {
     auto p_precise = RandomDataGenerator(rows, cols, 0).GenerateDMatrix();