1 / 13

mPL 5 Overview ISPD 2005 Placement Contest Entry

mPL 5 Overview ISPD 2005 Placement Contest Entry. Tony Chan 2 , Jason Cong 1 , Joe Shinnerl 1 , Kenton Sze 2 , Michalis Romesis 1 , Min Xie 1. University of California, Los Angeles. http://cadlab.cs.ucla.edu/~cong. cong@cs.ucla.edu. Given problem. Problem size decreases.

davisedward
Download Presentation

mPL 5 Overview ISPD 2005 Placement Contest Entry

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. mPL 5 Overview ISPD 2005 Placement Contest Entry Tony Chan2, Jason Cong1, Joe Shinnerl1, Kenton Sze2, Michalis Romesis1, Min Xie1 University of California, Los Angeles http://cadlab.cs.ucla.edu/~cong cong@cs.ucla.edu

  2. Given problem Problem size decreases Interpolation & Relaxation (optimization) Coarsening(Clustering) Multiscale Optimization Framework • Explores different scales of the solution space at different levels • Supports VERY FAST and SCALABLE methods • Supports inclusion of complicated objectives and constraints • Successful across MANY DIVERSE applications UCLA VLSICAD LAB

  3. Multiscale Placement Vocabulary • Coarsening: build a hierarchy of problem approximations by generalized recursive clustering (or partitioning) • Relaxation: improve the placement at each level by iterative optimization • Interpolation:transfer coarse-level solution to adjacent, finer level (generalized declustering) • Multilevel Flow: multiple traversals over multiple hierarchies (V-cycle variations) UCLA VLSICAD LAB

  4. Relative Wirelength A Brief History of mPL • mPL 1.1 • FC-Clustering • added partitioning to legalization • mPL 1.0 [ICCAD00] • Recursive ESC clustering • NLP at coarsest level • Goto discrete relaxation • Slot Assignment legalization • Domino detailed placement UNIFORM CELL SIZE • mPL 2.0 • RDFL relaxation • primal-dual netlist pruning • mPL 3.0 [ICCAD 03] • QRS relaxation • AMG interpolation • multiple V-cycles • cell-area fragmentation • mPL 4.0 • improved DP • better coarsening • backtracking V-cycle NON-UNIFORM CELL SIZE • mPL 5.0 • Multilevel Force-Directed 2002 2003 year 2000 2001 2004 UCLA VLSICAD LAB

  5. Merge each vertex with its “best” neighbor Merged Nets Coarsening by Recursive Aggregation • First-Choice Clustering (hMetis [Karypis 1999]). UCLA VLSICAD LAB

  6. Weighted Interpolation (Generalized Declustering) • Transfer a partial solution from a coarser level to its adjacent finer level • Place a component ( ) at the weighted average of the positions of the clusters containing its neighbors Place representative components Place others by weighted interpolation UCLA VLSICAD LAB

  7. mPL5 Generalized Force-Directed Placement • Smooth the density constraints by solving a Helmholtz Equation: • Assume Neumann boundary conditions: forces pointing outside the chip boundary are zero: • Log-sum-exp smooth approximation to half-perimeter wirelength [Naylor 2001; Kahng and Wang 2004]: UCLA VLSICAD LAB

  8. mPL5 Nonlinear-Programing Solution • Using the Uzawa algorithm to solve the above nonlinear constrained minimization problem, we iteratively solve • No matrix storage and no second derivatives are computed. • Separate updates for separate lagrange multipliers; the bin-density constraints are NOT lumped together • Use multilevel approach to speed-up computation and better quality UCLA VLSICAD LAB

  9. Geometric based FC clustering Iterated Multilevel Flow Make use of placement solution from 1st V-cycle First Choice (FC) clustering UCLA VLSICAD LAB

  10. Legalization • The Patoma flow for guaranteed legalizability (ASPDAC 2005) is not needed for the ISPD05 contest benchmarks… • …because they have so much white space. • Work on Patoma-enabled mPL (“mPL6”) for large-scale, high-utilization cases ( even above 99%!) continues. • In the presence of many fixed macros, we sometimes reduce HPWL more than 30% (!) simply by assigning cells to obstacle-free subregions before placing them. • Can apply LP/network-flow techniques. • Preserve the global placement as much as possible (tetris-like) UCLA VLSICAD LAB

  11. Important command-line options for ISPD 2005 suite • -shrinkRegion, -shrinkRatio • Usually get shorter wirelength by packing cells • Can pack cells to left, right, or center to a user-specified level of utilization • -addDummyCells • The best configuration may be highly nonuniform • Adding unconnected “dummy” filler cells enables mPL to find good nonuniform placements • -legalRegion • Divide the chip into rectangular regions defined by fixed-block boundaries. • Assign cells to subregions first, then legalize separately within the subregions. • -GFD::Gamma • Can fine tune the rate at which the weight of spreading forces is increased relative to wirelength forces UCLA VLSICAD LAB

  12. Improvements by parameter tuning UCLA VLSICAD LAB

  13. Acknowledgments • Collaborators • Chin-Chih Chang, Cadence • Tim Kong, Magma • Yuchun (Clara) Ma, UCLA CS • Xin Yuan, IBM • Sponsors • SRC Contracts 99-TJ-686 and 2003-TJ-1091 • NSF Awards CCF-0430077 & CCR-9901153 • ONR Contract N00014-03-1-0888 UCLA VLSICAD LAB

More Related