Specialized systems are

Inevitable

Already the norm

Practical

Heterogeneous parallelism

4

Virtual Worlds Personal Robotics Datainformatics

ScientificEngineering

Applications

Heterogeneous

Hardware

New Arch.

Programming Heterogeneous Parallelism

MPIHadoop

VerilogVHDL

CUDAOpenCL

ThreadsOpenMP

Architecture Specific Programming Models

High Performance But Low Productivity

Virtual Worlds Personal Robotics Datainformatics

ScientificEngineering

Applications

Heterogeneous

Hardware

New Arch.

Scala Jav

aPython

Ruby

C++ Clojur

e

Not enough semantic knowledge to compile

automatically

No restrictions

General-Purpose Languages

Make Specialized PL Practical

Inevitable

Already the norm

Practical

Domain Specific Languages Domain Specific Languages (DSLs)

Programming language with restricted expressiveness for a particular domain

High-level, usually declarative, and deterministic

Most popular parallel PL

Most Science apps start with this PL

Enabled GPU innovation

Virtual Worlds Personal Robotics Datainformatics

ScientificEngineering

Physics(Liszt)

Data Analytics (OptiQL)

Graph Alg.(Green Marl)

Machine Learning(OptiML)

Statistics(R)

Applications

Domain- Specific

Languages

Heterogeneous

Hardware

DSLCompiler

DSLCompiler

DSLCompiler

DSLCompiler

DSLCompiler

New Arch.

High Performance Domain-Specific Languages

HP DSLs Will Be PracticalEmbedded DSLs

Flexible embedding languages: Scala, F#Metaprogramming (aka staging)Blurring line between libraries and languages

DSL FrameworksSupport domain specific optimizationsSupport parallelism and locality optimizations Multiple backendse.g. Spoofax, Rose, Delite, Terra/Lua