Les cours en ligne
Institut d'Informatique
|
Chaire Francqui [INFOCHFR]
cours(s) enregistré(s) en 2003-2004:
| | Date d'enregistrement | Professeur | Système utilisé | Protection par mot de passe | Taille | Télécharger | | 19/03/2004 | Exterieur |  | NON | 36,07 Mo |  | | Remarque: Lecture 1: Out-of-order processors (1/2) In order to understand how to influence performance, size and power, one has to understand how modern deeply pipelined architectures actually work. This lesson will cover the following topics: deeply pipelined superscalar architectures, predicated execution, speculative execution (branch prediction, value prediction), multithreaded execution.
| | 19/03/2004 | Exterieur | | NON | 33,43 Mo | | | Remarque: Lecture 1: Out-of-order processors (2/2) In order to understand how to influence performance, size and power, one has to understand how modern deeply pipelined architectures actually work. This lesson will cover the following topics: deeply pipelined superscalar architectures, predicated execution, speculative execution (branch prediction, value prediction), multithreaded execution.
| | 26/03/2004 | Exterieur | | NON | 37,04 Mo | | | Remarque: Lecture 2: Memory hierarchies (1/2) The memory hierarchy plays an essential role in the performance and power consumption of applications. It is paramount that one understands the interaction between the different levels of the memory hierarchy in order to be able to fully exploit them. This lesson will cover the following topics: advanced cache structures (victim caches, stream buffers), multilevel cache hierarchies, advanced cache indexing techniques, randomizing cache functions, skewed associative caches, trace caches, interaction between virtual memory and caches, TLB-structures.
| | 26/03/2004 | Exterieur | | NON | 4,437 Mo | | | Remarque: Lecture 2: Memory hierarchies (2/2) The memory hierarchy plays an essential role in the performance and power consumption of applications. It is paramount that one understands the interaction between the different levels of the memory hierarchy in order to be able to fully exploit them. This lesson will cover the following topics: advanced cache structures (victim caches, stream buffers), multilevel cache hierarchies, advanced cache indexing techniques, randomizing cache functions, skewed associative caches, trace caches, interaction between virtual memory and caches, TLB-structures.
| | 02/04/2004 | Exterieur | | NON | 76,31 Mo | | | Remarque: Lecture 3: Optimizations for performance (1/2) Most compilers optimize programs for performance. Modern architectures need optimizations that maximally exploit the available resources. This lesson will cover the following topics: loop unrolling schemes, software pipelining, data layout for optimal cache usage, global program optimizations, the impact of the operating system, real-time processing.
| | 02/04/2004 | Exterieur | | NON | 64,09 Mo | | | Remarque: Lecture 3: Optimizations for performance (2/2) Most compilers optimize programs for performance. Modern architectures need optimizations that maximally exploit the available resources. This lesson will cover the following topics: loop unrolling schemes, software pipelining, data layout for optimal cache usage, global program optimizations, the impact of the operating system, real-time processing.
| | 23/04/2004 | Exterieur | | NON | 10,95 Mo | | Remarque: Lecture 4: Optimizations for size (2/2) Modern software development introduces lots of overhead in the executable. Whole program analysis at link-time or at post-link-time can be used to significantly reduce the size of aprogram. Reductions up to 50% and more have been reported. This lesson will cover the following topics: program compaction, program compression, program extraction, grammar-based compression, mixed width instruction sets, link-time optimization techniques.
La première partie de ce cours est indisponible dû à des problèmes techniques. Veuillez nous en excuser.
| | 30/04/2004 | Exterieur | | NON | 40,29 Mo | | Remarque: Lecture 5: Optimizations for power (1/2) For mobile applications power is a hard constraint. The minimal requirement is a minimum 16h autonomy, and a maximum 8h regeneration cycle per day. Battery technology can help to reach this goal, but also architectural and software optimizations. This lesson will cover the following topics: power save modes in processors, clock gating, software optimizations for power reduction.
La seconde partie de ce cours est indisponible dû à des problèmes techniques. Veuillez nous en excuser.
| | 07/05/2004 | Exterieur | | NON | 13,07 Mo | | | Remarque: Lecture 6: Program execution modeling (1/2) Processors are complex systems with a very complex dynamic behavior. Simulating is no longer an option to study processor performance due to the very long simulation time (one second of execution time requires about one day of simulation time). Statistical and analytical models can help us in understanding the behavior of a processor, and are essential in the design of new processors, and in advanced load balancers for non-uniform computing environments (performance estimation on a given platform). This lesson will cover the following topics: statistical modeling of program executions, analytical modeling of program executions, synthetic programs and synthetic traces, trace sampling, the cold start problem, the impact of input on program behavior.
| | 07/05/2004 | Exterieur | | NON | 9,265 Mo | | | Remarque: Lecture 6: Program execution modeling (2/2) Processors are complex systems with a very complex dynamic behavior. Simulating is no longer an option to study processor performance due to the very long simulation time (one second of execution time requires about one day of simulation time). Statistical and analytical models can help us in understanding the behavior of a processor, and are essential in the design of new processors, and in advanced load balancers for non-uniform computing environments (performance estimation on a given platform). This lesson will cover the following topics: statistical modeling of program executions, analytical modeling of program executions, synthetic programs and synthetic traces, trace sampling, the cold start problem, the impact of input on program behavior.
|
|
