1. Yingmin Li Ting Yan Qi Zhao
OOO vs. EPIC
Yingmin Li
Ting Yan
Qi Zhao

2. Outline
“Advantages” of EPIC
Critique
Conclusion

3. EPIC: Main Idea “Smart compiler, dumb machine” Finding parallelism
Processor  compiler
Software/hardware synergy
Processor design
Avoid complexity and difficulty
ILP, SMT & CMP

4. EPIC: Predication In OOO: dynamic branch prediction.
Larger basic blocks.
Control dep.  Data dep.
Eliminate misprediction & penalties.

5. EPIC: Speculation OOO: dynamic hardware
Data speculation & control speculation
Bigger window
Reduce impact of memory latencies

6. EPIC: Large Register Set
OOO: register renaming.
Easier to design than reg. Renaming.
“Real” registers benefits some apps.
Encryption alg., Numerical alg.
Avoid loss of invisible registers.
Interruptions in OOO.

7. EPIC: Unique Features Register Stack Engine (RSE).
To deal with call/ return costs.
Seems an unlimited stack of phys. Reg.
Rotating register file.
Software pipelining.
Multiple loops at the same time.

8. Function Call Register saving/restoring Register file Processor?
Compiler?
Register file
Expensive
Always idle

9. Predication
Computation of the branch condition is on the critical path
Increase ICache footprint
Half of the functional units effectively used if both “then” and “else” are scheduled
Hard to implement out-of-order with full predication

10. Predication
To compute if (a) x = t+1:

11. Control Speculation
Why not just use prefetch which will not cause unexpected exception?
Technique to exploit control speculation such as superblock increase code length

12. Control prediction

13. Data Speculation Moving a load above a possibly conflicting store
An advanced load and a checking load (IA64)
A run-time predictor

14. Data speculation

15. Software Pipelining For high performance technical computing
High trip-count loops
For commercial applications
Low trip-count loops

16. EPIC: at least not a breakthrough
Design Object of EPIC:
Moving hardware complexity to compiler

17. EPIC: at least not a breakthrough
The failure of EPIC:
The compiling technique used for EPIC almost also apply well to OOO
Hardware simplicity is not so obvious to offset EPIC’s overhead
Without dynamic information, compiler essentially can’t do sth well enough

18. The tragedy of cycle time
Why no obvious improvement in cycle time
mechanisms like RSA increase die complexity
Compare and dependent branch in one cycle
Predicted execution dependent on the existence of many function units

19. Dynamic path length: hey, IA64, you wasted too much here
Speculation
Half of the predicted instructions discarded
Restricted bundling
One base register
No sign-extended loads
No integer multiply or divide in general register

20. CPI No dynamic prediction
Longer source code (more GR, Predicate register, template bit, restricted bundling, recovery code) is burdensome for instruction fetching
Recovery code may induce ICache pollution or just a page-fault