(Downloads - 0)
For more info about our services contact : help@bestpfe.com
Table of contents
I State of the art
2 Geophysics and seismic applications
2.1 Introduction to seismic exploration
2.1.1 Seismic acquisition
2.1.2 Seismic processing
2.1.3 Seismic interpretation
2.2 Seismic migrations and Reverse Time Migration (RTM)
2.2.1 Description and overview of migration methods
2.2.2 Reverse Time Migration
2.3 Numerical methods for the wave propagation phenomena
2.3.1 The wave equation
2.3.1.1 Seismic waves and propagation media
2.3.1.2 The elastic wave equation
2.3.1.3 The acoustic wave equation
2.3.2 Numerical methods for wave propagation
2.3.2.1 Integral methods
2.3.2.2 Asymptotic methods
2.3.2.3 Direct methods
2.3.2.3.1 Pseudo-Spectral Methods
2.3.2.3.2 Finite Difference Methods
2.3.2.3.3 Finite Element Methods
2.3.3 Application to the acoustic wave equation
2.3.3.1 Numerical approximation
2.3.3.2 Stability analysis and CFL
2.3.3.3 Boundary conditions
3 High performance computing
3.1 Overview of HPC hardware architectures
3.1.1 Central Processing Unit: more and more cores
3.1.2 Hardware accelerators: the other chips for computing
3.1.3 Towards the fusion of CPUs and accelerators: the emergence of the Accelerated Processing Unit
3.2 Programming models in HPC
3.2.1 Dedicated programming languages for HPC
3.2.1.1 Overview
3.2.1.2 The OpenCL programming model
3.2.2 Directive-based compilers and language extensions
3.3 Power consumption in HPC and the power wall
4 Overview of accelerated seismic applications
4.1 Stencil computations
4.2 Reverse time migration
4.2.1 Evolution of RTM algorithms
4.2.2 Wave-field reconstruction methods
4.2.2.1 Re-computation of the forward wavefield
4.2.2.2 Storing all the forward wavefield
4.2.2.3 Selective wavefield storage (linear checkpointing)
4.2.2.4 Checkpointing
4.2.2.5 Boundaries storage
4.2.2.6 Random boundary condition
4.2.3 RTM on multi-cores and hardware accelerators
4.2.3.1 RTM on multi-core CPUs
4.2.3.2 RTM on GPUs
4.2.3.3 RTM on other accelerators
4.3 Close to seismics workflows
5 Thesis position and contributions
5.1 Position of the study
5.2 Contributions
5.3 Hardware and seismic material configurations
5.3.1 The hardware configuration
5.3.2 The numerical configurations of the seismic materials
5.3.2.1 The seismic source
5.3.2.2 The velocity model and the compute grids
6 Evaluation of the Accelerated Processing Unit (APU)
6.1 Data placement strategies
6.2 Applicative benchmarks
6.2.1 Matrix multiplication
6.2.1.1 Implementation details
6.2.1.2 Devices performance
6.2.1.3 Impact of data placement strategies on performance
6.2.1.4 Performance comparison
6.2.2 Finite difference stencil
6.2.2.1 Implementation details
6.2.2.2 Devices performance
6.2.2.3 Impact of data placement strategies on performance
6.2.2.4 Performance comparison
6.3 Power consumption aware benchmarks
6.3.1 Power measurement tutorial
6.3.1.1 Metrics for power efficiency
6.3.1.2 Proposed methodology
6.3.1.3 Hardware configuration
6.3.1.4 Choice of applications and benchmarks
6.3.2 Power efficiency of the applicative benchmarks
6.4 Hybrid utilization of the APU: finite difference stencil as an example
6.4.1 Hybrid strategy for the APU
6.4.2 Deployment on CPU or on integrated GPU
6.4.3 Hybrid deployment
6.5 Directive based programming on the APU: finite difference stencil as an example
6.5.1 OpenACC implementation details
6.5.2 OpenACC performance numbers and comparison with OpenCL
7 Seismic applications on one compute node
7.1 Seismic modeling
7.1.1 Description of the algorithm
7.1.2 Accelerating the seismic modeling using OpenCL
7.1.3 Performance and power efficiency
7.1.4 OpenACC evaluation and comparison with OpenCL
7.2 Seismic migration
7.2.1 Description of the algorithm
7.2.2 Accelerating the seismic migration using OpenCL
7.2.3 Performance and power efficiency
7.2.4 OpenACC evaluation and comparison with OpenCL
7.3 Conclusion
8 Large scale seismic applications on CPU/APU/GPU clusters
8.1 Large scale considerations
8.1.1 Domain decomposition
8.1.2 Boundary conditions
8.2 Seismic modeling
8.2.1 Deployment on CPU clusters: performance issues and proposed solutions
8.2.1.1 Implementation details
8.2.1.2 Communications and related issues
8.2.1.3 Load balancing
8.2.1.4 Communication-computation overlap
8.2.1.4.1 Problems of non-blocking MPI communications
8.2.1.4.2 Proposed solutions
8.2.1.4.3 Performance results
8.2.2 Deployment on hardware accelerators
8.2.2.1 Implementation details
8.2.2.2 Performance results
8.2.2.2.1 Strong scaling tests
8.2.2.2.2 Weak scaling tests
8.3 Seismic migration
8.3.1 Deployment on CPU clusters
8.3.1.1 Implementation details
8.3.1.2 Performance results
8.3.1.2.1 Strong scaling tests
8.3.1.2.2 Weak scaling tests
8.3.2 Deployment on hardware accelerators
8.3.2.1 Implementation details
8.3.2.2 Performance results
8.3.2.2.1 Strong scaling tests
8.3.2.2.2 Weak scaling tests
8.3.3 Performance comparison
8.3.3.1 Comparison based on measured results
8.3.3.2 Comparison based on performance projection
8.4 Conclusion
9 Conclusions and perspectives
9.1 Conclusions
9.2 Perspectives
