[incubator-mxnet] branch master updated: [BUGFIX] Fix CI slowdown issue after removing 3rdparty/openmp (#20367)

[akarbown]

This is an automated email from the ASF dual-hosted git repository.

akarbown pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git


The following commit(s) were added to refs/heads/master by this push:
     new 5dbba54  [BUGFIX] Fix CI slowdown issue after removing 3rdparty/openmp 
(#20367)
5dbba54 is described below

commit 5dbba54812742f4d6f9d83b1f7bae88084b69d75
Author: bgawrych <bartlomiej.gawr...@intel.com>
AuthorDate: Tue Jun 22 09:22:20 2021 +0200

    [BUGFIX] Fix CI slowdown issue after removing 3rdparty/openmp (#20367)
    
    * Fix CI slowdown issue after removing 3rdparty/openmp
    
    * Fix sanity
---
 src/operator/tensor/broadcast_reduce-inl.h | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

diff --git a/src/operator/tensor/broadcast_reduce-inl.h 
b/src/operator/tensor/broadcast_reduce-inl.h
index 1907c02..13fadbf 100644
--- a/src/operator/tensor/broadcast_reduce-inl.h
+++ b/src/operator/tensor/broadcast_reduce-inl.h
@@ -360,10 +360,17 @@ 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);
+    }
   }
 }