Add missing cudaFree to emitted pool_base_ptr memory buffer (#640)

* Added corresponding cudaFree to the cudaMalloc for the cuda pool_base_ptr memory buffer.

* Check for temporary buffer allocation prior to freeing. Add null check on cudaFree.

src/ngraph/runtime/gpu/gpu_external_function.cpp

@@ -711,6 +711,10 @@ using namespace std;
                 }
             }
         }
+        if (temporaries_used)
+        {
+            writer << "ngraph::runtime::gpu::free_gpu_buffer(pool_base_ptr);\n";
+        }
 
         writer.indent--;
         // End generated function

src/ngraph/runtime/gpu/gpu_util.cpp

@@ -54,6 +54,14 @@ void* runtime::gpu::create_gpu_buffer(size_t buffer_size)
     return allocated_buffer_pool;
 }
 
+void runtime::gpu::free_gpu_buffer(void* buffer)
+{
+    if (buffer)
+    {
+        cudaFree(buffer);
+    }
+}
+
 void runtime::gpu::cuda_memcpyDtD(void* dst, void* src, size_t buffer_size)
 {
     cudaMemcpy(dst, src, buffer_size, cudaMemcpyDeviceToDevice);

src/ngraph/runtime/gpu/gpu_util.hpp

@@ -61,6 +61,7 @@ namespace ngraph
             void print_gpu_f32_tensor(void* p, size_t element_count, size_t element_size);
             void check_cuda_errors(CUresult err);
             void* create_gpu_buffer(size_t buffer_size);
+            void free_gpu_buffer(void* buffer);
             void cuda_memcpyDtD(void* dst, void* src, size_t buffer_size);
             void cuda_memcpyHtD(void* dst, void* src, size_t buffer_size);
             void cuda_memset(void* dst, int value, size_t buffer_size);