|  |  | @ -24,6 +24,12 @@ namespace paddle { | 
			
		
	
		
		
			
				
					
					|  |  |  | namespace memory { |  |  |  | namespace memory { | 
			
		
	
		
		
			
				
					
					|  |  |  | namespace allocation { |  |  |  | namespace allocation { | 
			
		
	
		
		
			
				
					
					|  |  |  | 
 |  |  |  | 
 | 
			
		
	
		
		
			
				
					
					|  |  |  |  |  |  |  | /**
 | 
			
		
	
		
		
			
				
					
					|  |  |  |  |  |  |  |  * CUDADeviceContextAllocator will allocate a CUDADeviceContextAllocation | 
			
		
	
		
		
			
				
					
					|  |  |  |  |  |  |  |  * after waiting for a self-created event on the default stream. It does so to | 
			
		
	
		
		
			
				
					
					|  |  |  |  |  |  |  |  * let the non-default stream be able to allocate GPU memory which will be | 
			
		
	
		
		
			
				
					
					|  |  |  |  |  |  |  |  * released by stream callback | 
			
		
	
		
		
			
				
					
					|  |  |  |  |  |  |  |  */ | 
			
		
	
		
		
			
				
					
					|  |  |  | class CUDADeviceContextAllocator : public Allocator { |  |  |  | class CUDADeviceContextAllocator : public Allocator { | 
			
		
	
		
		
			
				
					
					|  |  |  |  public: |  |  |  |  public: | 
			
		
	
		
		
			
				
					
					|  |  |  |   explicit CUDADeviceContextAllocator(platform::CUDAPlace place, |  |  |  |   explicit CUDADeviceContextAllocator(platform::CUDAPlace place, | 
			
		
	
	
		
		
			
				
					|  |  | 
 |