disable CUDA async

Author: akroviakov

examples/heterogeneous_demo copy.ipynb

@@ -44,7 +44,7 @@ class BufferProvider {
                                  const int8_t* host_ptr,
                                  const size_t num_bytes,
                                  const int device_id) const = 0;
-  virtual void synchronizeStream(const int device_id) const = 0;
+  virtual void synchronizeDeviceDataStream(const int device_id) const = 0;
   virtual void copyFromDevice(int8_t* host_ptr,
                               const int8_t* device_ptr,
                               const size_t num_bytes,

omniscidb/BufferProvider/BufferProvider.h

@@ -131,7 +131,7 @@ void CudaMgr::copyHostToDeviceAsync(int8_t* device_ptr,
   checkError(cuMemcpyHtoDAsync(
       reinterpret_cast<CUdeviceptr>(device_ptr), host_ptr, num_bytes, stream_));
 }
-void CudaMgr::synchronizeStream(const int device_num) {
+void CudaMgr::synchronizeDeviceDataStream(const int device_num) {
   setContext(device_num);
   checkError(cuStreamSynchronize(stream_));
 }

omniscidb/CudaMgr/CudaMgr.cpp

@@ -106,7 +106,7 @@ class CudaMgr : public GpuMgr {
                              const size_t num_bytes,
                              const int device_num) override;
 
-  void synchronizeStream(const int device_num) override;
+  void synchronizeDeviceDataStream(const int device_num) override;
 
   void copyDeviceToHost(int8_t* host_ptr,
                         const int8_t* device_ptr,
@@ -294,7 +294,7 @@ class CudaMgr : public GpuMgr {
   omnisci::DeviceGroup device_group_;
   std::vector<CUcontext> device_contexts_;
   mutable std::mutex device_cleanup_mutex_;
-  static constexpr bool async_data_load_available{true};
+  static constexpr bool async_data_load_available{false};
 };
 
 }  // Namespace CudaMgr_Namespace

omniscidb/CudaMgr/CudaMgr.h

@@ -50,7 +50,7 @@ void CudaMgr::copyHostToDeviceAsyncIfPossible(int8_t* device_ptr,
   CHECK(false);
 }
 
-void CudaMgr::synchronizeStream(const int device_num) {
+void CudaMgr::synchronizeDeviceDataStream(const int device_num) {
   CHECK(false);
 }
 

omniscidb/CudaMgr/CudaMgrNoCuda.cpp

@@ -86,5 +86,5 @@ void GpuAllocator::setDeviceMem(int8_t* device_ptr,
 }
 
 void GpuAllocator::sync(){
-  buffer_provider_->synchronizeStream(device_id_);
+  buffer_provider_->synchronizeDeviceDataStream(device_id_);
 }

omniscidb/DataMgr/Allocators/GpuAllocator.cpp

@@ -60,11 +60,11 @@ void DataMgrBufferProvider::copyToDeviceAsyncIfPossible(int8_t* device_ptr,
   gpu_mgr->copyHostToDeviceAsyncIfPossible(device_ptr, host_ptr, num_bytes, device_id);
 }
 
-void DataMgrBufferProvider::synchronizeStream(const int device_num) const {
+void DataMgrBufferProvider::synchronizeDeviceDataStream(const int device_num) const {
   CHECK(data_mgr_);
   const auto gpu_mgr = data_mgr_->getGpuMgr();
   CHECK(gpu_mgr);
-  gpu_mgr->synchronizeStream(device_num);
+  gpu_mgr->synchronizeDeviceDataStream(device_num);
 }
 
 void DataMgrBufferProvider::copyFromDevice(int8_t* host_ptr,

omniscidb/DataMgr/DataMgrBufferProvider.cpp

@@ -45,7 +45,7 @@ class DataMgrBufferProvider : public BufferProvider {
                          const int8_t* host_ptr,
                          const size_t num_bytes,
                          const int device_id) const override;
-  void synchronizeStream(const int device_id) const override;
+  void synchronizeDeviceDataStream(const int device_id) const override;
   void copyFromDevice(int8_t* host_ptr,
                       const int8_t* device_ptr,
                       const size_t num_bytes,

omniscidb/DataMgr/DataMgrBufferProvider.h

@@ -43,7 +43,7 @@ struct GpuMgr {
                                                const size_t num_bytes,
                                                const int device_num) = 0;
 
-  virtual void synchronizeStream(const int device_num) = 0;
+  virtual void synchronizeDeviceDataStream(const int device_num) = 0;
 
   virtual void copyDeviceToHost(int8_t* host_ptr,
                                 const int8_t* device_ptr,