Penn ESE535 Spring 2009 -- DeHon 1 ESE535: Electronic Design Automation Day 20: April 13, 2009 Placement II (Simulated Annealing)

Slides:



Advertisements
Similar presentations
Penn ESE 535 Spring DeHon 1 ESE535: Electronic Design Automation Day 12: February 23, 2011 Routing 2 (Pathfinder)
Advertisements

Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 23: April 10, 2013 Statistical Static Timing Analysis.
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 11: March 4, 2008 Placement (Intro, Constructive)
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 23: April 22, 2009 Statistical Static Timing Analysis.
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 9: February 20, 2008 Partitioning (Intro, KLFM)
Penn ESE Spring DeHon 1 ESE (ESE534): Computer Organization Day 15: March 12, 2007 Interconnect 3: Richness.
Placement 1 Outline Goal What is Placement? Why Placement?
EDA (CS286.5b) Day 5 Partitioning: Intro + KLFM. Today Partitioning –why important –practical attack –variations and issues.
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 21: April 15, 2009 Routing 1.
Penn ESE Fall DeHon 1 ESE (ESE534): Computer Organization Day 19: March 26, 2007 Retime 1: Transformations.
Reconfigurable Computing (EN2911X, Fall07)
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 7: February 11, 2008 Static Timing Analysis and Multi-Level Speedup.
Penn ESE Spring DeHon 1 ESE (ESE534): Computer Organization Day 11: February 14, 2007 Compute 1: LUTs.
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 22: April 11, 2011 Statistical Static Timing Analysis.
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 19: April 9, 2008 Routing 1.
Simulated Annealing 10/7/2005.
EDA (CS286.5b) Day 11 Scheduling (List, Force, Approximation) N.B. no class Thursday (FPGA) …
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 5: February 2, 2009 Architecture Synthesis (Provisioning, Allocation)
EDA (CS286.5b) Day 7 Placement (Simulated Annealing) Assignment #1 due Friday.
Penn ESE525 Spring DeHon 1 ESE535: Electronic Design Automation Day 10: February 18, 2009 Partitioning 2 (spectral, network flow)
CSE 144 Project Part 2. Overview Multiple rows Routing channel between rows Components of identical height but various width Goal: Implement a placement.
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 15: March 18, 2009 Static Timing Analysis and Multi-Level Speedup.
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 5: February 2, 2009 Architecture Synthesis (Provisioning, Allocation)
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 8: February 13, 2008 Retiming.
CSE 242A Integrated Circuit Layout Automation Lecture 5: Placement Winter 2009 Chung-Kuan Cheng.
Placement by Simulated Annealing. Simulated Annealing  Simulates annealing process for placement  Initial placement −Random positions  Perturb by block.
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 4: January 28, 2015 Partitioning (Intro, KLFM)
CALTECH CS137 Fall DeHon 1 CS137: Electronic Design Automation Day 17: November 11, 2005 Placement (Simulated Annealing…)
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 2: January 26, 2015 Covering Work preclass exercise.
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 9: February 16, 2015 Scheduling Variants and Approaches.
Massachusetts Institute of Technology 1 L14 – Physical Design Spring 2007 Ajay Joshi.
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 9: February 14, 2011 Placement (Intro, Constructive)
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 24: April 18, 2011 Covering and Retiming.
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 10: February 18, 2015 Architecture Synthesis (Provisioning, Allocation)
CALTECH CS137 Winter DeHon CS137: Electronic Design Automation Day 6: January 23, 2002 Partitioning (Intro, KLFM)
Penn ESE525 Spring DeHon 1 ESE535: Electronic Design Automation Day 6: February 4, 2014 Partitioning 2 (spectral, network flow)
CALTECH CS137 Winter DeHon CS137: Electronic Design Automation Day 10: February 6, 2002 Placement (Simulated Annealing…)
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 23: April 20, 2015 Static Timing Analysis and Multi-Level Speedup.
CALTECH CS137 Winter DeHon CS137: Electronic Design Automation Day 13: February 20, 2002 Routing 1.
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 13: March 3, 2015 Routing 1.
CALTECH CS137 Winter DeHon CS137: Electronic Design Automation Day 9: February 9, 2004 Partitioning (Intro, KLFM)
FPGA CAD 10-MAR-2003.
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 10: February 16, 2011 Placement II (Simulated Annealing)
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 6: January 30, 2013 Partitioning (Intro, KLFM)
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 20: April 4, 2011 Static Timing Analysis and Multi-Level Speedup.
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 15: March 13, 2013 High Level Synthesis II Dataflow Graph Sharing.
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 11: February 25, 2015 Placement (Intro, Constructive)
CS137 Electronic Design Automation Day 9: February 25, 2004 Placement II André DeHon, Michael Wrighton.
CALTECH CS137 Winter DeHon 1 CS137: Electronic Design Automation Day 8: January 27, 2006 Cellular Placement.
CALTECH CS137 Fall DeHon 1 CS137: Electronic Design Automation Day 21: November 28, 2005 Routing 1.
Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 25: April 17, 2013 Covering and Retiming.
Placement and Routing Algorithms. 2 FPGA Placement & Routing.
Local Search Algorithms
CS137: Electronic Design Automation
ESE535: Electronic Design Automation
ESE535: Electronic Design Automation
ESE535: Electronic Design Automation
ESE535: Electronic Design Automation
ESE534: Computer Organization
ESE535: Electronic Design Automation
ESE535: Electronic Design Automation
ESE535: Electronic Design Automation
ESE535: Electronic Design Automation
ESE535: Electronic Design Automation
ESE535: Electronic Design Automation
Xin-She Yang, Nature-Inspired Optimization Algorithms, Elsevier, 2014
ESE535: Electronic Design Automation
ESE535: Electronic Design Automation
ESE535: Electronic Design Automation
Reconfigurable Computing (EN2911X, Fall07)
Presentation transcript:

Penn ESE535 Spring DeHon 1 ESE535: Electronic Design Automation Day 20: April 13, 2009 Placement II (Simulated Annealing)

Penn ESE535 Spring DeHon 2 Today Placement Improving Quality –Cost functions –Avoiding local minima Technique: –Simulated Annealing Behavioral (C, MATLAB, …) RTL Gate Netlist Layout Masks Arch. Select Schedule FSM assign Two-level, Multilevel opt. Covering Retiming Placement Routing

Penn ESE535 Spring DeHon 3 Simulated Annealing Physically motivated approach Physical world has similar problems –objects/atoms seeking minimum cost arrangement –at high temperature (energy) can move around E.g. it melts –at low temperature, no free energy to move –cool quickly  freeze in defects (weak structure) glass –cool slowly  allow to find minimum cost crystal

Penn ESE535 Spring DeHon 4 Key Benefit Avoid Local Minima –Allowed to take locally non-improving moves in order to avoid being stuck

Penn ESE535 Spring DeHon 5 Simulated Annealing At high temperature can move around –not trapped to only make “improving” moves –free energy from “temperature” allows exploration of non-minimum states –avoid being trapped in local minima As temperature lowers –less energy available to take big, non-minimizing moves –more local / greedy moves

Penn ESE535 Spring DeHon 6 Design Optimization Components: 1.“Energy” (Cost) function to minimize –represent entire state, drives system forward 2.Moves –local rearrangement/transformation of solution 3.Cooling schedule –initial temperature –temperature steps (sequence) –time at each temperature

Penn ESE535 Spring DeHon 7 Basic Algorithm Sketch Pick an initial solution Set temperature (T) to initial value while (T>T min ) –for time at T pick a move at random compute  cost if less than zero, accept else if (RND<e -  cost/T ), accept –update T

Penn ESE535 Spring DeHon 8 Details Initial Temperature –T 0 =  avg/ln(P accept ) –e -  cost/T –e -  cost/T0 =e -  cost/(  avg/ln(Paccept)) –Average move  e ln(Paccept) Assume increasing cost is negative  avg Accepted with Probability P accept

Penn ESE535 Spring DeHon 9 Details Cooling schedule –fixed ratio: T= T (e.g. =0.85) –temperature dependent –function of both temperature and acceptance rate example to come Time at each temperature –fixed number of moves? –fixed number of rejected moves? –fixed fraction of rejected moves?

Penn ESE535 Spring DeHon 10 VPR Cooling Schedule Moves at Temperature = cN 4/3 Temperature Update –Tnew=Told×  –Idea: advance slowly in good  range –  is measured acceptance rate Betz, Rose, & Marquardt Kluwer 1999

Penn ESE535 Spring DeHon 11 Cost Function Can be very general –Combine area, timing, energy, routability… Desirable characteristics: 1.drive entire solution in right direction reward each good move 2.cheap to compute delta costs e.g. FM Ideally O(1)

Penn ESE535 Spring DeHon 12 Bad Cost Functions Not reward every move: –size < threshold ? –Anything using max channel width critical path delay Expensive update cost –rerun router on every move –rerun static timing analysis E.g. recalculate critical path delay

Penn ESE535 Spring DeHon 13 Example Cost Functions Total Wire Length –Linear, quadratic…  Bounding Box (semi-perimeter) –Surrogate for routed net length  (e channel_density ) –Dominate by largest density –Rewards improvement in non-maximum channel –But reward is larger for denser channels –Can be computed incrementally bb y bb x

Penn ESE535 Spring DeHon 14 VPR Wire Costs VPR Bounding Box Swartz, Betz, & Rose FPGA 1998 Original table: Cheng ICCAD 1994

Penn ESE535 Spring DeHon 15 VPR Timing Costs Criticality(e)=1-Slack(e)/Dmax TCost(e)=Delay(e)*Criticality(e) CriticalityExp Keep all edge delays in a table Recompute Net Criticality at each Temperature Marquardt, Betz, & Rose FPGA2000

Penn ESE535 Spring DeHon 16 VPR Balance Wire and Time Cost Marquardt, Betz, & Rose FPGA2000

Penn ESE535 Spring DeHon 17 Stale Criticality Criticality becomes stale during moves –Strategy of updating STA infrequently problematic for highly pipelined logic Eguro & Hauck DAC 2008

Penn ESE535 Spring DeHon 18 Initial Solution Spectral Placement Random Constructive Placement –Fast placers start at lower temperature; assume constructive got global right.

Penn ESE535 Spring DeHon 19 Moves Swap two cells –Within some distance limit? (ex. to come) swap regions –…rows, columns, subtrees, cluster rotate cell (when feasible) flip (mirror) cell permute cell inputs (equivalent inputs)

Penn ESE535 Spring DeHon 20 Legality Constraints Examples: –Limit on number of gates/cluster (position) –Limit on number of Inputs/cluster (region) Options: –Force all moves to be legal Force initial placement to be legal Illegal moves rejected –Allow illegal placement/moves Set cost function to make undesirable Make less desirable (more costly) over time

Penn ESE535 Spring DeHon 21 Basic Algorithm Sketch (review) Pick an initial solution Set temperature (T) to initial value while (T>T min ) –for time at T pick a move at random compute  cost if less than zero, accept else if (RND<e -  cost/T ), accept –update T

Penn ESE535 Spring DeHon 22 Variant: “Rejectionless” Order moves by cost –compare FM Pick random number first Use random to define range of move costs will currently accept Pick randomly within this range Idea: never pick a costly move which will be rejected

Penn ESE535 Spring DeHon 23 Simulated Annealing Theory If stay long enough at each cooling stage –will achieve tight error bound If cool long enough –will find optimum …but is it any less work than exhaustive exploration? –Good to have a continuum….

Penn ESE535 Spring DeHon 24 Practice Good results –ultimately, what most commercial tools use...what vpr uses… Slow convergence Tricky to pick schedules to accelerate convergence –Too slow  runs too long –Too fast  freezes prematurely  local min  low quality

Penn ESE535 Spring DeHon 25 Pragmatic Approach Good way to find out what optimization is possible –Run for long time and cool slowly –If can slow down cooling and get improvement Demonstration haven’t found optimum, yet Once know good result this way –Can try to accelerate convergence –w/out sacrificing quality

Penn ESE535 Spring DeHon 26 Range Limit Want to tune so accepting 44% of the moves – Lam and Delosme DAC 1988 VPR –Define Rlimit – defines maximum  x and  y accepted –Tune Rlimit to maintain acceptance rate –Rlimit new =Rlimit old ×(  )  is measured acceptance rate

Penn ESE535 Spring DeHon 27 Range Limiting? Eguro alternate [DAC 2005] – define P=D -M –Tune M to control 

Penn ESE535 Spring DeHon 28 Range Limiting Eurgo, Hauck, & Sharma DAC 2005 Distance is P swap =D -M M adjusted For 44% accept

Penn ESE535 Spring DeHon 29 Big Hammer Costly, but general Works for most all problems –(part, placement, route, retime, schedule…) Can have hybrid/mixed cost functions –as long as weight to single potential –(e.g. wire/time from VPR) With care, can attack multiple levels –place and route Ignores structure of problem –resignation to finding/understanding structure

Penn ESE535 Spring DeHon 30 Summary Simulated Annealing –use randomness to explore space –accept “bad” moves to avoid local minima –decrease tolerance over time General purpose solution –costly in runtime

Penn ESE535 Spring DeHon 31 Admin Reading for Monday online

Penn ESE535 Spring DeHon 32 Big Ideas: Use randomness to explore large (non- convex) space –Sample various parts of space –Avoid becoming trapped in local minimum Technique –Simulated Annealing