From 3d8d5b0f8568eca5d6977737511dbeeb97eddb72 Mon Sep 17 00:00:00 2001
From: igor_rukhovich <igor.rukhovich@intel.com>
Date: Thu, 19 Nov 2020 16:19:58 +0300
Subject: [PATCH 1/2] Improved InitSampling function speed by 2.12 times

---
 src/tree/updater_quantile_hist.cc | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/src/tree/updater_quantile_hist.cc b/src/tree/updater_quantile_hist.cc
index 46a82ea3887e..ee40e63b17bd 100644
--- a/src/tree/updater_quantile_hist.cc
+++ b/src/tree/updater_quantile_hist.cc
@@ -691,17 +691,18 @@ void QuantileHistMaker::Builder<GradientSumT>::InitSampling(const std::vector<Gr
     r = rnd;
   }
   const size_t discard_size = info.num_row_ / nthread;
+  uint32_t coin_flip_border = static_cast<uint32_t>(std::numeric_limits<uint32_t>::max()
+                                                    * param_.subsample);
   #pragma omp parallel num_threads(nthread)
   {
     const size_t tid = omp_get_thread_num();
     const size_t ibegin = tid * discard_size;
     const size_t iend = (tid == (nthread - 1)) ?
                         info.num_row_ : ibegin + discard_size;
-    std::bernoulli_distribution coin_flip(param_.subsample);
 
-    rnds[tid].discard(2*discard_size * tid);
+    rnds[tid].discard(discard_size * tid);
     for (size_t i = ibegin; i < iend; ++i) {
-      if (gpair[i].GetHess() >= 0.0f && coin_flip(rnds[tid])) {
+      if (gpair[i].GetHess() >= 0.0f && rnds[tid]() < coin_flip_border) {
         p_row_indices[ibegin + row_offsets[tid]++] = i;
       }
     }

From f797c7b4996f5af46ba9959773c2c6df6e6487b7 Mon Sep 17 00:00:00 2001
From: igor_rukhovich <igor.rukhovich@intel.com>
Date: Thu, 19 Nov 2020 23:09:51 +0300
Subject: [PATCH 2/2] Added explicit conversion

---
 src/tree/updater_quantile_hist.cc | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/tree/updater_quantile_hist.cc b/src/tree/updater_quantile_hist.cc
index ee40e63b17bd..a751673d5164 100644
--- a/src/tree/updater_quantile_hist.cc
+++ b/src/tree/updater_quantile_hist.cc
@@ -691,8 +691,8 @@ void QuantileHistMaker::Builder<GradientSumT>::InitSampling(const std::vector<Gr
     r = rnd;
   }
   const size_t discard_size = info.num_row_ / nthread;
-  uint32_t coin_flip_border = static_cast<uint32_t>(std::numeric_limits<uint32_t>::max()
-                                                    * param_.subsample);
+  auto upper_border = static_cast<float>(std::numeric_limits<uint32_t>::max());
+  uint32_t coin_flip_border = static_cast<uint32_t>(upper_border * param_.subsample);
   #pragma omp parallel num_threads(nthread)
   {
     const size_t tid = omp_get_thread_num();