fix ZeroDivisionError in utils.bottleneck (pytorch#11987)

Summary:
**ZeroDivisionError** occurs when `cuda_prof_exec_time` is small enough.
This situation is normal for a project that has little CUDA work.

Or someone does not make his work transferred to CUDA successfully. In this time he profiles the code, this error occurs.
Pull Request resolved: pytorch#11987

Differential Revision: D10488568

Pulled By: soumith

fbshipit-source-id: db8c1e9e88a00943c100958ebef41a1cb56e7e65

torch/utils/bottleneck/__main__.py

@@ -222,10 +222,12 @@ def main():
     # Print both the result of the CPU-mode and CUDA-mode autograd profilers
     # if their execution times are very different.
     cuda_prof_exec_time = cpu_time_total(autograd_prof_cuda)
-    cpu_prof_exec_time = cpu_time_total(autograd_prof_cpu)
-    pct_diff = cuda_prof_exec_time - cpu_prof_exec_time / cuda_prof_exec_time
-    if abs(pct_diff) > 0.05:
-        print_autograd_prof_summary(autograd_prof_cpu, 'CPU', autograd_prof_sortby, autograd_prof_topk)
+    if len(autograd_prof_cpu.function_events) > 0:
+        cpu_prof_exec_time = cpu_time_total(autograd_prof_cpu)
+        pct_diff = (cuda_prof_exec_time - cpu_prof_exec_time) / cuda_prof_exec_time
+        if abs(pct_diff) > 0.05:
+            print_autograd_prof_summary(autograd_prof_cpu, 'CPU', autograd_prof_sortby, autograd_prof_topk)
+
     print_autograd_prof_summary(autograd_prof_cuda, 'CUDA', autograd_prof_sortby, autograd_prof_topk)
 
 if __name__ == '__main__':