search

vetter / shoc

The SHOC Benchmark Suite
Other
247 stars 104 forks source link

OpenCL resources not released in MaxFlops #17

Closed speziale-ettore closed 11 years ago

speziale-ettore commented 11 years ago

Hi,

In the MaxFlops benchmark, when running more than one iteration, OpenCL events used to timing the MAD kernel are not freed. Here is the patch:

diff --git a/src/opencl/level0/MaxFlops.cpp b/src/opencl/level0/MaxFlops.cpp
index 0639d10..765c331 100644
--- a/src/opencl/level0/MaxFlops.cpp
+++ b/src/opencl/level0/MaxFlops.cpp
@@ -588,10 +588,11 @@ RunTest(cl_device_id id,
         CL_CHECK_ERROR(err);

-        // Event object for timing
-        Event evKernel_madd("madd");
         for (int passCounter=0; passCounter < npasses; passCounter++)
         {
+            // Event object for timing
+            Event evKernel_madd("madd");
+
             err = clEnqueueNDRangeKernel(queue, kernel_madd, 1, NULL,
                       &globalWorkSize, &localWorkSize,
                       0, NULL, &evKernel_madd.CLEvent());

Bye, Ettore Speziale