1. My major mantra: We Must Break the Layers

2. Algorithm Program ISA (Instruction Set Arch)‏ Microarchitecture Circuits Problem Electrons

3. IF we break the layers: Compiler, Microarchitecture –Multiple levels of cache –Block-structured ISA –Part by compiler, part by uarch –Fast track, slow track Algorithm, Compiler, Microarchitecture –X + superscalar – the Refrigerator –Niagara X / Pentium Y Microarchitecture, Circuits –Verification Hooks –Internal fault tolerance

4. …and More intelligently design our core (ILP is not dead ) Provide more than one interface Re-visit the Run-time system

5. Morphcore High performance out-of-order for serial code SMT in-order when throughput dominates Mickey mouse when neither is present Who decides which mode to operate in?

6. More than one interface Those who need the utmost in performance –and are willing to trade ease of use for extra performance –For them, we need to provide a more powerful engine (Heavyweight cores for ILP, Accelerators for specific tasks ) Those who just need to get their job done –and are content with trading performance for ease of use –For them, we need a software layer to bridge the multi-core That means –Chip designers understanding what algorithms need –Algorithms folks understanding what hardware provides –Knowledgeable Software folks bridging the two interfaces

7. The Run-time System: What? Part of the Operating System? Part of the Chip Design?

8. The Run-time System: Why? MorphCore (Khubaib, IEEE/ACM Micro 2012) Adaptive Processors, (Miftakhudinov, Micro 2012) Shared resources (Ebrahimi, IEEE/ACM ISCA 2012) Which thread to speed up (Joao, ISCA 2013)

9. The Run-time System: How?

10. Algorithm Program ISA (Instruction Set Arch)‏ Microarchitecture Circuits Problem Electrons

11. The Run-time System: How? Programmers insert pragmas in their code Compilers use it to insert procedure calls Serious on-chip monitoring provides inputs Run-time system decides how to utilize on-chip resources