Difference between revisions of "WG211/M3Danelutto"
m (1 revision) |
|||
| Line 2: | Line 2: | ||
| − | + | Marco Danelutto | |
| − | + | ||
| − | + | ||
| − | |||
| − | |||
Title: | Title: | ||
| Line 18: | Line 16: | ||
In this talk, we shortly outline classical and more untraditional skeletal system implementations and we show how layered skeletal system implementation can exploit both compile time and the run time techniques/methods and tools to achieve high efficiency in the execution of skeleton programs. <br> | In this talk, we shortly outline classical and more untraditional skeletal system implementations and we show how layered skeletal system implementation can exploit both compile time and the run time techniques/methods and tools to achieve high efficiency in the execution of skeleton programs. <br> | ||
In particular, we'll discuss the experience at the Pisa research group that used both implementation strategies (traditional, template based, and untraditional, based on macro data flow and source-to-source skeleton program transformations) in distinct skeleton based programming environments. | In particular, we'll discuss the experience at the Pisa research group that used both implementation strategies (traditional, template based, and untraditional, based on macro data flow and source-to-source skeleton program transformations) in distinct skeleton based programming environments. | ||
| − | |||
| − | |||
| − | |||
| − | |||
* [[Media:DaneluttoSlidesPDF.pdf | DaneluttoSlidesPDF.pdf ]]: Slides of the talk (January 25, 2006 - Dagstuhl) | * [[Media:DaneluttoSlidesPDF.pdf | DaneluttoSlidesPDF.pdf ]]: Slides of the talk (January 25, 2006 - Dagstuhl) | ||
Revision as of 14:13, 16 December 2011
Marco Danelutto
Title:
Generative aspects in skeletal systems
Speaker: Marco Danelutto
Abstract:
Algorithmical skeleton based environments have been around since the '90. Historically, they have been implemented exploiting template libraries hosting known, parametric process networks implementing each one of the skeletons provided by the system. This technology has been used to implement both compiler based and library based skeletal systems. Later on, new implementation techniques have been developed that interpret the skeleton program provided by the user as meta information to be processed when deriving/generating the actual, low level parallel code to be eventually executed on the target architecture.
<p>
In this talk, we shortly outline classical and more untraditional skeletal system implementations and we show how layered skeletal system implementation can exploit both compile time and the run time techniques/methods and tools to achieve high efficiency in the execution of skeleton programs.
In particular, we'll discuss the experience at the Pisa research group that used both implementation strategies (traditional, template based, and untraditional, based on macro data flow and source-to-source skeleton program transformations) in distinct skeleton based programming environments.
- DaneluttoSlidesPDF.pdf : Slides of the talk (January 25, 2006 - Dagstuhl)
