From 4a54845678cb2a4e7bf28e573bcdb20b8e228342 Mon Sep 17 00:00:00 2001
From: Stephen Tiedemann <stephen.tiedemann@sony.com>
Date: Wed, 12 Jun 2019 11:04:48 +0200
Subject: [PATCH] Fix cuda kernel block calculation for zero size.

---
 include/nbla/cuda/common.hpp | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/include/nbla/cuda/common.hpp b/include/nbla/cuda/common.hpp
index 8a5e3c0c1..8b34e7949 100644
--- a/include/nbla/cuda/common.hpp
+++ b/include/nbla/cuda/common.hpp
@@ -180,6 +180,8 @@ enum {
     The kernel is assumed to contain a grid-strided loop.
  */
 inline int cuda_get_blocks_by_size(int size) {
+  if (size == 0)
+    return 0;
   const int blocks = NBLA_CUDA_GET_BLOCKS(size);
   const int inkernel_loop = NBLA_CEIL_INT_DIV(blocks, NBLA_CUDA_MAX_BLOCKS);
   const int total_blocks = NBLA_CEIL_INT_DIV(blocks, inkernel_loop);