Merge pull request JuliaGPU#205 from JuliaGPU/tb/query

Add query for stream and event.

src/events.jl

@@ -51,6 +51,23 @@ Waits for an event to complete.
 """
 synchronize(e::CuEvent) = cuEventSynchronize(e)
 
+"""
+ query(e::CuEvent)
+
+Return `false` if there is outstanding work preceding the most recent
+call to `record(e)` and `true` if all captured work has been completed.
+"""
+function query(e::CuEvent)
+ res = unsafe_cuEventQuery(e)
+ if res === ERROR_NOT_READY
+ return false
+ elseif res === SUCCESS
+ return true
+ else
+ throw_api_error(res)
+ end
+end
+
 """
  wait(e::CuEvent, stream=CuDefaultStream())
 

src/stream.jl

@@ -50,3 +50,20 @@ Return the default stream.
 Wait until a stream's tasks are completed.
 """
 synchronize(s::CuStream) = cuStreamSynchronize(s)
+
+"""
+ query(s::CuStream)
+
+Return `false` if a stream is busy (has task running or queued)
+and `true` if that stream is free.
+"""
+function query(s::CuStream)
+ res = unsafe_cuStreamQuery(s)
+ if res === ERROR_NOT_READY
+ return false
+ elseif res === SUCCESS
+ return true
+ else
+ throw_api_error(res)
+ end
+end

test/events.jl

@@ -29,6 +29,11 @@ CuEvent(CUDAdrv.EVENT_BLOCKING_SYNC | CUDAdrv.EVENT_DISABLE_TIMING)
  synchronize()
 end
 
+@testset "event query" begin
+ event = CuEvent()
+ @test CUDAdrv.query(event) == true
+end
+
 @testset "elapsed stream" begin
  stream = CuStream()
  @test (CUDAdrv.@elapsed stream begin end) > 0

test/stream.jl

@@ -3,6 +3,8 @@
 let
  s = CuStream()
  synchronize(s)
+ @test CUDAdrv.query(s) == true
+
  let s2 = CuStream()
  @test s != s2
  end