address review comments

Author: leofang

cuda_core/cuda/core/experimental/_event.py

@@ -48,7 +48,8 @@ class Event:
 
     Events can be used to monitor device's progress, query completion
     of work up to event's record, help establish dependencies
-    between GPU work submissions, and record the elapsed time on GPU:
+    between GPU work submissions, and record the elapsed time (in milliseconds)
+    on GPU:
 
     .. code-block:: python
 
@@ -58,14 +59,14 @@ class Event:
         # ... run some GPU works ...
         e2 = s.record(options={"enable_timing": True})
         e2.sync()
-        print(f"time = {e2 - e1}")
+        print(f"time = {e2 - e1} milliseconds")
 
         # Or, if events are already created:
         s.record(e1)
         # ... run some more GPU works ...
         s.record(e2)
         e2.sync()
-        print(f"time = {e2 - e1}")
+        print(f"time = {e2 - e1} milliseconds")
 
     Directly creating an :obj:`~_event.Event` is not supported due to ambiguity,
     and they should instead be created through a :obj:`~_stream.Stream` object.
@@ -120,7 +121,7 @@ def __rsub__(self, other):
         return NotImplemented
 
     def __sub__(self, other):
-        # return self - other
+        # return self - other (in milliseconds)
         timing = handle_return(driver.cuEventElapsedTime(other.handle, self.handle))
         return timing
 

cuda_core/tests/test_event.py

@@ -11,23 +11,28 @@
 import pytest
 
 from cuda.core.experimental import Device, EventOptions
+from cuda.core.experimental._utils import CUDAError
 
 
 @pytest.mark.parametrize("enable_timing", [True, False, None])
 def test_timing(init_cuda, enable_timing):
     options = EventOptions(enable_timing=enable_timing)
     stream = Device().create_stream()
-    n_seconds = 0.5
+    delay_seconds = 0.5
     e1 = stream.record(options=options)
-    time.sleep(n_seconds)
+    time.sleep(delay_seconds)
     e2 = stream.record(options=options)
+    e2.sync()
     for e in (e1, e2):
-        assert e.is_timing_disabled == (not enable_timing if enable_timing is not None else True)
+        assert e.is_timing_disabled == (True if enable_timing is None else not enable_timing)
     if enable_timing:
-        e2.sync()
-        elapsed_time = e2 - e1
-        assert isinstance(elapsed_time, float)
-        assert n_seconds * 1000 <= elapsed_time < n_seconds * 1000 + 2  # tolerance 2 ms
+        elapsed_time_ms = e2 - e1
+        assert isinstance(elapsed_time_ms, float)
+        assert delay_seconds * 1000 <= elapsed_time_ms < delay_seconds * 1000 + 2  # tolerance 2 ms
+    else:
+        with pytest.raises(CUDAError) as e:
+            elapsed_time_ms = e2 - e1
+            assert "CUDA_ERROR_INVALID_HANDLE" in str(e)
 
 
 def test_is_sync_busy_waited(init_cuda):