fix max_block_size in train states (fix #3570) (#3575)

* remove max_block_size_ in train states (fix #3570)

* avoid zero elements per row

* add min constraint for min_block_size_

include/LightGBM/train_share_states.h

@@ -57,7 +57,7 @@ class MultiValBinWrapper {
       n_data_block_ = 1;
       data_block_size_ = num_data;
       Threading::BlockInfo<data_size_t>(num_threads_, num_data, min_block_size_,
-                                        max_block_size_, &n_data_block_, &data_block_size_);
+                                        &n_data_block_, &data_block_size_);
       ResizeHistBuf(hist_buf, cur_multi_val_bin, origin_hist_data);
       OMP_INIT_EX();
       #pragma omp parallel for schedule(static) num_threads(num_threads_)
@@ -137,7 +137,6 @@ class MultiValBinWrapper {
   const std::vector<int> feature_groups_contained_;
 
   int num_threads_;
-  int max_block_size_;
   int num_bin_;
   int num_bin_aligned_;
   int n_data_block_;

include/LightGBM/utils/threading.h

@@ -40,24 +40,6 @@ class Threading {
     }
   }
 
-  template <typename INDEX_T>
-  static inline void BlockInfo(int num_threads, INDEX_T cnt,
-                               INDEX_T min_cnt_per_block, INDEX_T max_cnt_per_block,
-                               int* out_nblock, INDEX_T* block_size) {
-    CHECK(max_cnt_per_block >= min_cnt_per_block);
-    *out_nblock = std::min<int>(
-        num_threads,
-        static_cast<int>((cnt + min_cnt_per_block - 1) / min_cnt_per_block));
-    *out_nblock = std::max<int>(
-      *out_nblock,
-      static_cast<int>((cnt + max_cnt_per_block - 1) / max_cnt_per_block));
-    if (*out_nblock > 1) {
-      *block_size = SIZE_ALIGNED((cnt + (*out_nblock) - 1) / (*out_nblock));
-    } else {
-      *block_size = cnt;
-    }
-  }
-
   template <typename INDEX_T>
   static inline void BlockInfoForceSize(int num_threads, INDEX_T cnt,
                                         INDEX_T min_cnt_per_block,

src/io/train_share_states.cpp

@@ -12,7 +12,6 @@ MultiValBinWrapper::MultiValBinWrapper(MultiValBin* bin, data_size_t num_data,
   const std::vector<int>& feature_groups_contained):
     feature_groups_contained_(feature_groups_contained) {
   num_threads_ = OMP_NUM_THREADS();
-  max_block_size_ = num_data;
   num_data_ = num_data;
   multi_val_bin_.reset(bin);
   if (bin == nullptr) {
@@ -39,8 +38,10 @@ void MultiValBinWrapper::InitTrain(const std::vector<int>& group_feature_start,
   if (cur_multi_val_bin != nullptr) {
     num_bin_ = cur_multi_val_bin->num_bin();
     num_bin_aligned_ = (num_bin_ + kAlignedSize - 1) / kAlignedSize * kAlignedSize;
+    auto num_element_per_row = cur_multi_val_bin->num_element_per_row();
     min_block_size_ = std::min<int>(static_cast<int>(0.3f * num_bin_ /
-      cur_multi_val_bin->num_element_per_row()) + 1, 1024);
+      (num_element_per_row + kZeroThreshold)) + 1, 1024);
+    min_block_size_ = std::max<int>(min_block_size_, 32);
   }
 }