–Autotuning will use CRESTA’s autotuning framework
–EU FP7 Network project: Collaborative Research into Exascale Software Tools and Applications
–A way to address inherent complexity in current and future systems
–Allows us to make automated choices over how an application is built and deployed
–Choices are for example:
–algorithm
–source (optional branches, data flow, loop transformations)
–compilation
–application launch
–Our approach is to specifically define a DSL to expose these choices and describe a framework to work within. Target parallel aspects.
–We did not have enough effort to delve into some areas:
machine learning, models, compiler-based approaches, ML database
–Autotuning will use CRESTA’s autotuning framework –EU FP7 Network project: Collaborative Research into Exascale Software Tools and Applications –A way to address inherent complexity in current and future systems –Allows us to make automated choices over how an application is built and deployed
–Choices are for example: –algorithm –source (optional branches, data flow, loop transformations) –compilation –application launch –Our approach is to specifically define a DSL to expose these choices and describe a framework to work within. Target parallel aspects. –We did not have enough effort to delve into some areas: machine learning, models, compiler-based approaches, ML database