1. ALCHEMY Architectures, Languages and Compilers to Harness the End of Moore Years  INRIA project (INRIA Futurs, Saclay)  Main research focus of Alchemy: Long-Term performance scalability of architectures  4 faculty A. Cohen (CR INRIA) C. Eisenbeis (DR INRIA) F. Gruau (MC LRI) O. Temam (DR INRIA)  3 postdocs & visitors  10 PhDs

2. Compiler optimizations Compiler optimizations Architecture optimizations Architecture optimizations Manual optimizations (decision tree) Manual optimizations (decision tree) Software optimizations compatible with complex processor architectures & applications Software optimizations compatible with complex processor architectures & applications Scalable processor architectures Scalable processor architectures Performance  Usage ? Performance  Usage ? Iterative optimization (environment & COP) Iterative optimization (environment & COP) Compositions of transformations (polyhedral model) Compositions of transformations (polyhedral model) Influence of data sets Influence of data sets Algorithm selection Algorithm selection Architecture- inspired software optimizations (VHC) Architecture- inspired software optimizations (VHC) Methodology (comparison) Methodology (comparison) Simulators: Development & Execution (MicroLib) Simulators: Development & Execution (MicroLib) Language+Architecture (domain-specific, e.g., video processing) Language+Architecture (domain-specific, e.g., video processing) Language+Architecture (general-purpose) Language+Architecture (general-purpose) Customization (functions) Customization (functions) ALCHEMY Processor simulation Processor simulation Complex systems Complex systems BLOB Computing (model) BLOB Computing (model) New technologies ? New technologies ? Bio-inspired systems Bio-inspired systems

3. Iterative/Adaptive Optimization & Center for Program Tuning (COP)  Iterative/Adaptive optimization  rapidly adjusts to a large spectrum of complex processor architectures  how to use it in practice ? Workstations New execution Program Learning across iterations Program transformations End of execution COP University of Edinburgh

4. Symbiotic Processing  The clock race is ending (see Intel)  Easily translate on-chip space into performance  Focus on on-chip parallelism  Constraints Application  let user extract parallelism effortlessly through proper programming paradigm Architecture  no central control (too large space), local control only  avoid complex architectures  Spatial computing break down program into independent/interacting objects only local actions architecture manages objects architecture exploits parallelism (object execution & replication) depending on available resources/space  Application to future SMTs & CMPs Pentium 4

5. MicroLib http://www.microlib.org  Change the way people do research in the domain: facilitate exchange, reuse and comparison of ideas  A library of simulator components (cache, BP, FU…) for complex processors  Open library (open-source, anyone can participate)  Already heavily used processor architecture components  cache, branch prediction, scheduling… research modules  12 data cache mechanisms (MICRO 2004) full processor simulators  PowerPC 750 (~15% accurate) 2000+ downloads in 18 months  OoOSysC (generic superscalar)  simple MIPS 2000-like RISC  Promoting library HiPEAC (simulation framework) team up with competition: Liberty environment at Princeton University organizer of the next international micro-architecture challenge on data caches