add wait event for kernel

src/operators/kernel/cl/batchnorm_kernel.cpp

@@ -86,8 +86,11 @@ void BatchNormKernel<GPU_CL, float>::Compute(
   clSetKernelArg(kernel, 4, sizeof(cl_mem), &new_bias);
   clSetKernelArg(kernel, 5, sizeof(cl_mem), &out);
 
+  cl_event out_event = param.OutputY()->GetClEvent();
+  cl_event wait_event = param.InputX()->GetClEvent();
   clEnqueueNDRangeKernel(this->cl_helper_.CLCommandQueue(), kernel, 3, NULL,
-                         default_work_size.data(), NULL, 0, NULL, NULL);
+                         default_work_size.data(), NULL, 1, &wait_event,
+                         &out_event);
 }
 
 template class BatchNormKernel<GPU_CL, float>;

src/operators/kernel/cl/fetch_kernel.cpp

@@ -73,8 +73,9 @@ void FetchKernel<GPU_CL, float>::Compute(const FetchParam<GPU_CL> &param) {
     clSetKernelArg(kernel, 6, sizeof(int), &size_batch);
   }
 
+  cl_event wait_event = param.InputX()->GetClEvent();
   clEnqueueNDRangeKernel(this->cl_helper_.CLCommandQueue(), kernel, 3, NULL,
-                         default_work_size.data(), NULL, 0, NULL, NULL);
+                         default_work_size.data(), NULL, 1, &wait_event, NULL);
 
   memcpy(out->data<float>(), out_cl_tensor.Data<float>(), out->memory_size());
 }

src/operators/kernel/cl/pool_kernel.cpp

@@ -63,8 +63,10 @@ void PoolKernel<GPU_CL, float>::Compute(const PoolParam<GPU_CL> &param) {
   clSetKernelArg(kernel, 10, sizeof(cl_mem), &input);
   clSetKernelArg(kernel, 11, sizeof(cl_mem), &out);
 
+  cl_event out_event = param.Output()->GetClEvent();
+  cl_event wait_event = param.Input()->GetClEvent();
   clEnqueueNDRangeKernel(this->cl_helper_.CLCommandQueue(), kernel, 3, NULL,
-                         default_work_size.data(), NULL, 0, NULL, NULL);
+                         default_work_size.data(), NULL, 1, &wait_event, &out_event);
 }
 
 template class PoolKernel<GPU_CL, float>;