From 2e98479d5d4480e7d6c4a6a1bc9150816dec2ec2 Mon Sep 17 00:00:00 2001
From: "B. Gawrych" <bartlomiej.gawrych@intel.com>
Date: Mon, 21 Jun 2021 09:06:54 +0200
Subject: [PATCH 1/2] Fix CI slowdown issue after removing 3rdparty/openmp

---
 src/operator/tensor/broadcast_reduce-inl.h | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)

diff --git a/src/operator/tensor/broadcast_reduce-inl.h b/src/operator/tensor/broadcast_reduce-inl.h
index 1907c02897c9..b03b3b0d3bca 100644
--- a/src/operator/tensor/broadcast_reduce-inl.h
+++ b/src/operator/tensor/broadcast_reduce-inl.h
@@ -360,11 +360,19 @@ void seq_reduce_compute(const size_t N, const size_t M, const bool addto,
                         const Shape<ndim> sshape, const Shape<ndim> rshape,
                         const Shape<ndim> rstride) {
   const int thread_count = engine::OpenMP::Get()->GetRecommendedOMPThreadCount();
-  #pragma omp parallel for num_threads(thread_count) if (N >= thread_count)
-  for (index_t idx = 0; idx < static_cast<index_t>(N); ++idx) {
-    seq_reduce_assign<Reducer, ndim, AType, DType, OType, OP, IndexOP>
-        (idx, M, addto, big, small, bshape, sshape, rshape, rstride, N < thread_count);
+  if (N >= thread_count) {
+    #pragma omp parallel for num_threads(thread_count)
+    for (index_t idx = 0; idx < static_cast<index_t>(N); ++idx) {
+      seq_reduce_assign<Reducer, ndim, AType, DType, OType, OP, IndexOP>
+          (idx, M, addto, big, small, bshape, sshape, rshape, rstride, false);
+    }
+  } else {
+    for (index_t idx = 0; idx < static_cast<index_t>(N); ++idx) {
+      seq_reduce_assign<Reducer, ndim, AType, DType, OType, OP, IndexOP>
+          (idx, M, addto, big, small, bshape, sshape, rshape, rstride, true);
+    }
   }
+
 }
 
 template <typename Reducer, int ndim, typename DType, typename OP>

From 28ef3045abebc77495cb1caf94f32da0735ed2ad Mon Sep 17 00:00:00 2001
From: "B. Gawrych" <bartlomiej.gawrych@intel.com>
Date: Mon, 21 Jun 2021 11:21:08 +0200
Subject: [PATCH 2/2]  Fix sanity

---
 src/operator/tensor/broadcast_reduce-inl.h | 1 -
 1 file changed, 1 deletion(-)

diff --git a/src/operator/tensor/broadcast_reduce-inl.h b/src/operator/tensor/broadcast_reduce-inl.h
index b03b3b0d3bca..13fadbf41918 100644
--- a/src/operator/tensor/broadcast_reduce-inl.h
+++ b/src/operator/tensor/broadcast_reduce-inl.h
@@ -372,7 +372,6 @@ void seq_reduce_compute(const size_t N, const size_t M, const bool addto,
           (idx, M, addto, big, small, bshape, sshape, rshape, rstride, true);
     }
   }
-
 }
 
 template <typename Reducer, int ndim, typename DType, typename OP>