Merge pull request #168 from sony/fix/20190605_zero_div_in_cuda_block…

…s_for_size_0 Fix cuda kernel block calculation for zero size.
sony · Jul 1, 2019 · 15c1e9a · 15c1e9a
2 parents e2e031e + 4a54845
commit 15c1e9a
Showing 1 changed file with 2 additions and 0 deletions.
diff --git 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);