Added launch bounds to the reduce kernels (apache#16397)

* Added launch bounds to the reduce_kernel_M1 * Trigger CI * Reretrigger the CI
yajiedesign · Nov 6, 2019 · 3dc5ce9 · 3dc5ce9
1 parent 955b5fe
commit 3dc5ce9
Showing 1 changed file with 2 additions and 0 deletions.
diff --git a/src/operator/tensor/broadcast_reduce-inl.cuh b/src/operator/tensor/broadcast_reduce-inl.cuh
@@ -262,6 +262,7 @@ __global__ void reduce_lines_kernel(const int N, const int M, const bool addto,
 }
 
 template<typename Reducer, int ndim, typename AType, typename DType, typename OType, typename OP>
+__launch_bounds__(kMaxThreadsPerBlock)
 __global__ void reduce_kernel_M1(const int N, const bool addto,
                                 const DType* __restrict big, OType *small, const Shape<ndim> bshape,
                                 const Shape<ndim> sshape) {
@@ -277,6 +278,7 @@ __global__ void reduce_kernel_M1(const int N, const bool addto,
 }
 
 template<typename Reducer, int ndim, typename DType, typename OP1, typename OP2>
+__launch_bounds__(kMaxThreadsPerBlock)
 __global__ void reduce_kernel_M1(const int N, const bool addto,
                                  const DType* __restrict big,
                                  const DType* __restrict lhs,