diff --git a/ompi/mca/coll/cuda/coll_cuda.h b/ompi/mca/coll/cuda/coll_cuda.h
index afedc632ee5..4b3ecc647eb 100644
--- a/ompi/mca/coll/cuda/coll_cuda.h
+++ b/ompi/mca/coll/cuda/coll_cuda.h
@@ -54,7 +54,7 @@ int mca_coll_cuda_reduce(const void *sbuf, void *rbuf, int count,
                          struct ompi_communicator_t *comm,
                          mca_coll_base_module_t *module);
 
-int mca_coll_cuda_reduce_local(const void *sbuf, void *rbuf, size_t count,
+int mca_coll_cuda_reduce_local(const void *sbuf, void *rbuf, int count,
                                struct ompi_datatype_t *dtype,
                                struct ompi_op_t *op,
                                mca_coll_base_module_t *module);
diff --git a/ompi/mca/coll/cuda/coll_cuda_reduce.c b/ompi/mca/coll/cuda/coll_cuda_reduce.c
index 7743a078743..e165a1d9bb8 100644
--- a/ompi/mca/coll/cuda/coll_cuda_reduce.c
+++ b/ompi/mca/coll/cuda/coll_cuda_reduce.c
@@ -85,7 +85,7 @@ mca_coll_cuda_reduce(const void *sbuf, void *rbuf, int count,
 }
 
 int
-mca_coll_cuda_reduce_local(const void *sbuf, void *rbuf, size_t count,
+mca_coll_cuda_reduce_local(const void *sbuf, void *rbuf, int count,
                            struct ompi_datatype_t *dtype,
                            struct ompi_op_t *op,
                            mca_coll_base_module_t *module)