fix possible bug related num_threads (#2876)

* only one fix

* add more

* add more

include/LightGBM/utils/common.h

@@ -735,7 +735,7 @@ static void ParallelSort(_RanIt _First, _RanIt _Last, _Pr _Pred, _VTRanIt*) {
   size_t inner_size = (len + num_threads - 1) / num_threads;
   inner_size = std::max(inner_size, kMinInnerLen);
   num_threads = static_cast<int>((len + inner_size - 1) / inner_size);
-  #pragma omp parallel for schedule(static, 1)
+#pragma omp parallel for schedule(static, 1)
   for (int i = 0; i < num_threads; ++i) {
     size_t left = inner_size*i;
     size_t right = left + inner_size;

include/LightGBM/utils/threading.h

@@ -118,7 +118,7 @@ class ParallelPartitionRunner {
     }
 
     OMP_INIT_EX();
-#pragma omp parallel for schedule(static, 1)
+#pragma omp parallel for schedule(static, 1) num_threads(num_threads_)
     for (int i = 0; i < nblock; ++i) {
       OMP_LOOP_EX_BEGIN();
       INDEX_T cur_start = i * inner_size;
@@ -156,7 +156,7 @@ class ParallelPartitionRunner {
     data_size_t left_cnt = left_write_pos_[nblock - 1] + left_cnts_[nblock - 1];
 
     auto right_start = out + left_cnt;
-#pragma omp parallel for schedule(static)
+#pragma omp parallel for schedule(static, 1) num_threads(num_threads_)
     for (int i = 0; i < nblock; ++i) {
       std::copy_n(left_.data() + offsets_[i], left_cnts_[i],
                   out + left_write_pos_[i]);

src/io/dataset.cpp

@@ -512,7 +512,7 @@ MultiValBin* Dataset::GetMultiBinFromSparseFeatures() const {
   std::vector<uint32_t> most_freq_bins;
   double sum_sparse_rate = 0;
   for (int i = 0; i < num_feature; ++i) {
-#pragma omp parallel for schedule(static)
+#pragma omp parallel for schedule(static, 1)
     for (int tid = 0; tid < num_threads; ++tid) {
       iters[tid].emplace_back(
           feature_groups_[multi_group_id]->SubFeatureIterator(i));
@@ -556,7 +556,7 @@ MultiValBin* Dataset::GetMultiBinFromAllFeatures() const {
           num_total_bin -= 1;
         }
         offsets.push_back(num_total_bin);
-#pragma omp parallel for schedule(static)
+#pragma omp parallel for schedule(static, 1)
         for (int tid = 0; tid < num_threads; ++tid) {
           iters[tid].emplace_back(
               feature_groups_[gid]->SubFeatureIterator(fid));
@@ -1228,7 +1228,7 @@ void Dataset::ConstructHistogramsMultiVal(const data_size_t* data_indices,
     hist_data = share_state->TempBuf();
   }
   OMP_INIT_EX();
-#pragma omp parallel for schedule(static)
+#pragma omp parallel for schedule(static, 1) num_threads(share_state->num_threads)
   for (int tid = 0; tid < n_data_block; ++tid) {
     OMP_LOOP_EX_BEGIN();
     data_size_t start = tid * data_block_size;
@@ -1261,7 +1261,7 @@ void Dataset::ConstructHistogramsMultiVal(const data_size_t* data_indices,
   int bin_block_size = num_bin;
   Threading::BlockInfo<data_size_t>(share_state->num_threads, num_bin, 512, &n_bin_block,
                                     &bin_block_size);
-#pragma omp parallel for schedule(static)
+#pragma omp parallel for schedule(static, 1) num_threads(share_state->num_threads)
   for (int t = 0; t < n_bin_block; ++t) {
     const int start = t * bin_block_size;
     const int end = std::min(start + bin_block_size, num_bin);
@@ -1333,7 +1333,7 @@ void Dataset::ConstructHistogramsInner(
       }
     }
     OMP_INIT_EX();
-#pragma omp parallel for schedule(static)
+#pragma omp parallel for schedule(static) num_threads(share_state->num_threads)
     for (int gi = 0; gi < num_used_dense_group; ++gi) {
       OMP_LOOP_EX_BEGIN();
       int group = used_dense_group[gi];

src/treelearner/serial_tree_learner.cpp

@@ -352,7 +352,7 @@ void SerialTreeLearner::FindBestSplitsFromHistograms(
   std::vector<int8_t> larger_node_used_features = col_sampler_.GetByNode();
   OMP_INIT_EX();
 // find splits
-#pragma omp parallel for schedule(static)
+#pragma omp parallel for schedule(static) num_threads(share_state_->num_threads)
   for (int feature_index = 0; feature_index < num_features_; ++feature_index) {
     OMP_LOOP_EX_BEGIN();
     if (!is_feature_used[feature_index]) {

src/treelearner/voting_parallel_tree_learner.cpp

@@ -358,7 +358,7 @@ void VotingParallelTreeLearner<TREELEARNER_T>::FindBestSplitsFromHistograms(cons
   // find best split from local aggregated histograms
 
   OMP_INIT_EX();
-  #pragma omp parallel for schedule(static)
+#pragma omp parallel for schedule(static) num_threads(this->share_state_->num_threads)
   for (int feature_index = 0; feature_index < this->num_features_; ++feature_index) {
     OMP_LOOP_EX_BEGIN();
     const int tid = omp_get_thread_num();