System partitioning. System functionality is implemented on system components ASICs, processors, memories, buses

Transcription

1 System partitioning System functionality is implemented on system components ASICs, processors, memories, buses Two design tasks: Allocate system components or ASIC constraints Partition functionality among components Constraints Cost, performance, size, power Partitioning is a central system design task 103 of 14

2 Outline Structural vs. functional partitioning Natural vs. executable language specifications Basic partitioning issues and algorithms Functional partitioning techniques for hardware Hardware/software partitioning Functional partitioning techniques for software Exploring tradeoffs with functional partitioning System partitioning 104 of 14

3 Structural vs. functional partitioning Structural: Implement structure, then partition Functional: Partition function, then implement Enables better size/performance tradeoffs Uses fewer objects, better for algorithms/humans Permits hardware/software solutions But, it s harder than graph partitioning System partitioning 105 of 14

4 Natural vs. executable language specifications Alternative methods for specifying functionality Natural languages common in practice Executable languages becoming popular Automated estimation/partitioning explores solutions Early verification reduces costly late changes Precision eases integration System partitioning 106 of 14

5 Basic partitioning issues Specification abstraction level Granularity Metrics and estimations Partitioning algorithms Objective and closeness functions System component allocation Output System partitioning 107 of 14

6 Basic partitioning issues (cont.) Specification-abstraction level: input definition Just indicating the language is insufficient Abstraction-level indicates amount of design already done e.g. task DFG, tasks, CDFG, FSMD Granularity: specification size in each object Fine granularity yields more possible designs Coarse granularity better for computation, designer interaction e.g. tasks, procedures, statement blocks, statements Component allocation: types and numbers e.g. ASICs, processors, memories, buses Output: format and uses e.g. new specification, hints to synthesis tool System partitioning 108 of 14

7 Basic partitioning issues (cont.) Metrics and estimations: "good" partition attributes e.g. cost, speed, power, size, pins, testability, reliability Estimates derived from quick, rough implementation Speed and accuracy are competing goals of estimation Objective and closeness functions Combines multiple metric values Closeness used for grouping before complete partition Weighted sum common e.g. 1 3! System partitioning 109 of 14

8 Basic partitioning issues (cont.) Algorithms: control strategies seeking best partition Constructive creates partition Iterative improves partition Key is to escape local minimum Cost A B Number of moves System partitioning 110 of 14

9 Typical partitioning-system configuration User interface Input Model Output Algorithms Estimators Objective function Design feedback System partitioning 111 of 14

10 Basic partitioning algorithms Clustering and multi-stage clustering [Joh67, LT91] Group migration (a.k.a. min-cut or Kernighan/Lin) [KL70, FM8] Ratio cut [KC91] Simulated annealing [KGV83] Genetic evolution Integer linear programming System partitioning 11 of 14

11 Hierarchical clustering Constructive algorithm using closeness metrics Overview Groups closest objects Recomputes closenesses Repeats until termination condition met Cluster tree maintains history of merges Cutline across the tree defines a partition System partitioning 113 of 14

12 "# $# &' % & $# $( & $#( ' $( % & $1 Hierarchical clustering algorithm /* Initialize each object as a group */ for each loop "# % $# end loop /* Compute closenesses between objects */ for each loop for each loop ComputeCloseness( ) end loop end loop % )#+* ( $( $#-, */ /* Merge closest objects and recompute closenesses while not Terminate( ) loop FindClosestObjects( % $( $#, ). &, &0/ $#/ for each loop ComputeCloseness( % )#( 1 * end loop end loop ) $1 $#(, return & System partitioning 114 of 14

13 Hierarchical clustering example o o 10 o o4 o o1 o1 0 o3 o o3 o o o4 o4 o4 o 1 Avg(10,10) = 10 Avg(15,5) = 0 o 1 o o 3 o 4 o 1 o o 3 o 4 o 1 o o 3 o 4 o 1 o o 3 o 4 (a) (b) (c) (d) System partitioning 115 of 14

14 Simulated annealing Iterative algorithm modeled after physical annealing process Overview Starts with initial partition and temperature Slowly decreases temperature For each temperature, generates random moves Accepts any move that improves cost Accepts some bad moves, less likely at low temperatures Results and complexity depend on temperature decrease rate System partitioning 116 of 14

15 ; :98 ; 3 :98 :98 D 3 ; 7 3 :98 H 3 F B > < 5 7 Simulated annealing algorithm initial temperature ) while not Frozen loop while not Equilibrium loop 7 Objfct( ) Move( Objfct( >@? 3 = < 43 >@? 4= < 43 ; >? 4= < 43 ; ) :98 A >@? 3 4= < 43 :98 7 B1)) then Random( CB :98 ) if (Accept( >@? 4= < 43 ; >@? 4= < 43 :98 7 end if end loop ) DecreaseTemp( end loop G LMNO L KJI 1 G 65 3 CB F 89: E where: System partitioning 117 of 14

16 Functional partitioning for hardware: BUD Goal: incorporate area/time into synthesis [MK90] Clusters CDFG operations into datapath modules Closeness metrics: Interconnecting wires Concurrency Shared hardware Each clustering corresponds to an allocation/scheduling Selects clustering with best area/time System partitioning 118 of 14

17 BUD example start (bit widths = 4) a b + = x cond x := a + b; if (a = b) c := ((x y) < z); cond 0 1 x y z < c = < finish (a) (b) (c) System partitioning 119 of 14

18 BUD example (cont.) +. = < Avg(.38,0) =.19 Avg(0,.4) =.1 + =< AVG(.19,.1) = =< + < = + < = + < = (a) Clusters Chip area A Expected cycle time T Objfct = AxT + =< +, =< +, =, < +,, =, < (best) (b) Chip Controller (c) + < = 3 clusters System partitioning 10 of 14

19 Functional partitioning for hardware: Aparty Extends BUD clustering to multiple stages [LT91] Different closeness metrics for each stage Closeness metrics: Control transfer reduction Data transfer reduction Hardware sharing System partitioning 11 of 14

20 Aparty example o o o 1 4 o 3 o 3 1 o o 3 o 1 o 4 o 3 o o o o1 o o3 o (a) (b) (c) System partitioning 1 of 14

21 Hardware/software partitioning Combined hardware/software systems are common Software is cheap, modiable,and quick to design Hardware is fast Special algorithms are needed to favor software Proposed algorithms Greedy [GD9] Hill climbing [EHB94] Binary-constraint search with hill climbing [VGG93] System partitioning 13 of 14

22 Functional partitioning for systems: Vulcan, Cosyma Vulcan [GD90]I Partitions CDFG operations among hardware only Group migration and simulated annealing algorithms Vulcan II [GD93] Partitions operations among hardware/software Architecture: processor, hardware, memory, bus All communication through memory Uses greedy algorithm, extracts behaviors from hardware Cosyma [EHB94] Partitions statement blocks among hardware/software Architecture: processor, hardware, memory, bus Simulated annealing, extracts behaviors from software System partitioning 14 of 14

23 Functional partitioning for systems: SpecSyn Solves three partitioning problems Behaviors to processors/asics Variables to memories Communication channels to buses Uses fast incremental-update estimators Covers both hardware and hardware/software partitioning [GVN94, VG9] System partitioning 15 of 14

24 P P P P P P P Q Exploring tradeoffs with functional partitioning chipset1 chipset chipset3 Each line represents a different vendor s chip set Each point represents an allocation and partition performance (microseconds) D Many designs quickly examined C A B cost (dollars) System partitioning 16 of 14

25 Summary Partitioning heavily influences design quality Functional partitioning is necessary Executable specification enables: Automation Exploration Documentation Variety of algorithms exist Variety of techniques exist for different applications System partitioning 17 of 14

26 Future directions Metrics from real design to guide partitioning Comparison of functional partitioning algorithms Impact of metric selections and orderings Impact of of granularity on partition quality Exploitation of regularity in partitioning System partitioning 18 of 14