C-C++ Multicore Application Programming @ MdS

Objectives
The purpose of this training is to provide a pedagogical presentation of C-C++ shared  memory  multicore application programming.  These  programming lan- guages provide  access to a substantial number  of multithreading programming environments and libraries,  like the native threads libraries (Pthreads in Linux- Unix, Windows threads) or basic, general purpose libraries like Boost, and high level programming environments like OpenMP  or the Intel Threading Building Blocks environment.  While the emphasis is put in OpenMP  as a widely adopted standard, a broader  vision of multithreading is useful to develop a clear vision of the strategic options open to developers.
 
The pedagogical  objectives of this training are:
 
• To provide  a unified presentation of the basic concepts of multithreading and  shared  memory  programming, underlining  the universality  of these concepts  and  discussing  the way in which  they  are  declined  in different programming environments.

 
• To  assess  the focus,  the strengths and  the eventual weaknesses  of the different  programming environment,  insisting  on  their  complementarity as well as on their  interoperability.  Examples  are  given on the usage  of Pthreads based  tools in an OpenMP  environment, in order  to cope with parallel  contexts that are not easily handled  in OpenMP.
 
• To  enhance  programmer’s insights on  threads,   not   multithreading lan- guages, by focusing on parallel  and concurrent patterns that occur in real applications, and  by  discussing  useful tricks, potential pitfalls and  best practices that are largely independent of the underlying  programming en- vironment.
 
 
Content
 
• Introduction to multithreading, overview of present hardaware and  soft- ware technologies
 
• Introducing threads, the relevance of multithreading
 
• Creating and managing  threads
 
•  Thread safe programming
 
• Concurrent access to shared  data: the need of mutual exclusion
 
• Event synchronization of threads
 
• Memory  coherence  and  memory  consistency issues.  The  role of memory consistency models.
 
• Pthread based library  of high level synchronization tools.
 
• Thread pools  as  high  level  thread management  utilities.    Task  centric programming versus thread centric programming.
 
• Complete overview of OpenMP, including all the new features of the latest
4.0 release.  In depth discussion of task centric programming in OpenMP.
 
• Full OpenMP  examples in data parallel  codes, and in pipelining  threads.
 
• Coments  on  the impact  and  relevance  of the Intel  Threading Building
Blocks (TBB)  library.
 
 
Each  one of this subjects is illustrated with an  important number  of ad  hoc examples  exhibiting pitfalls and  best practices.  A library  of high level utilities that simplify  the programming of delicate  concurrency  patterns (like  control parallelism  and  pipelines,  or point to point synchronizations among individual threads) developped  by  the author, is made  available.    Several  real  complete applications involving a substantial number  of delicate programming issues are discussed in detail.

Maison de la Simulation