| The reverse time migration(RTM)algorithm has unparalleled advantages in the traditional imaging algorithm.Because of the rid of the limitation of the formation dip,it has obvious advantages in the scene of complex geological model imaging.However,while achieving high-precision imaging,the algorithm brings great challenges to storage and communication efficiency,which has also become the main bottleneck that RTM cannot be widely used in the industry.In recent years,with the continuous expansion of exploration scale expand,the bottleneck is increasingly prominent.In this context,the RTM algorithm often introduces additional I/O operations due to lack of storage resources,and also generates extremely high communication delays between the CPU and GPU.Therefore,alleviating the limitations of the RTM algorithm in storage and communication has become an urgent problem to be solved.In this paper,a research on the storage and communication optimization of the RTM algorithm is carried out for the above problems.The main work is as follows:(1)A method of using data compression technology to alleviate the storage bottleneck of RTM is proposed.Compress the seismic data that occupies a large space in the RTM algorithm,and perform lossless decompression when the data involves subsequent calculation operations,effectively reducing the demand for storage resources during the operation of the algorithm and avoiding the lack of memory space.Additional I/O operations;compress the data that needs to participate in the I/O transmission task,reduce the I/O delay by reducing the transmission amount,and improve the overall execution efficiency of the reverse time offset algorithm.(2)A method for efficient communication between CPU and GPU with the help of pinned memory is proposed.For the seismic data involved in communication operations in the RTM algorithm,the data with larger data scale and less transmission frequency is placed in the fixed memory of the CPU side,and when it needs to be transmitted to the GPU side to participate in the calculation of the kernel function,the DMA The direct memory mapping method informs the GPU of the address of the data,which effectively reduces the delay caused by data communication tasks.(3)A task concurrent execution strategy is proposed to hide the running delay of each task to the greatest extent.The RTM algorithm is divided into multiple computing and communication sub-tasks according to stages,and the data dependencies between tasks are analyzed,and the concurrent execution optimization strategy is implemented for the tasks that are independent of each other and have different resource requirements,and overlap some tasks.In order to improve the overall operation efficiency of the algorithm. |
